This "controversy" had been getting me hackles up for some time until I finally I posted an email-disguised-rant to the XLR8YourMac site (GREAT SITE) from which I will steal and expand the content below for your amusement.  I will also be providing theory "and handwaiving" as to why observed results seem to differ from the predicted ones.  Well, I didn't have any early predictions, I was just tuning.

Nothing about this is meant to disparage XLR8YourMac.  The site is asking for feeback to answer the question and a site like that is where the action is happening.
 
Getting right to the point.

I keep reading various postings and discussions describing how some person used the PCI QE mod and reports it helps, but some other person(s) NOT USING THE SAME SYSTEMS FOR THE SAME USES IN THE SAME WAY responds with: Your wrong, It just can't help! and they refuse to believe that the first person is NOT an idiot.  This is amazingly rude and insulting.  The people insitsing it can't work have arguments consisting of some valid technical points, plud largely hinging on the relatively meaningless fact that Apple didn't include it on for reasons.

Yep, people installing OS X on unspported hardware and roaming mod sites are (as least partly) arguing the PCI Quartz Extreme is bad because Apple didn't support... but Apple had good reasons.
 
Lets take a look at well it works, what problems it causes, and some reasons.  Lets actually USE reason and BE Reasonable, but moody...that's it...reasonable and moody... and start with...
 
Apple also didn't support ANY of the first generation PC machine with OS X for reasons, but some people still find them usable running OS X nonethteless, and I believe them. 

Apple also doesn't provide GUI access to a variety of features that they do provice CLI controls for...for reasons,  didn't give the OS X client a GUI option for formating case sensitive drives ... for reasons, and provides a VERY limited Spot Light interface...for reasons, and...for reasons...has elected to prevent spot light from finding hidden files basically ever.

Apple is VERY sensitive to the needs of less technical users and has a tendency to remove or hide options that require more technical skill to evaluate and use, but THAT IS EXACTLY WHAT A SITE LIKE XLR8YourMac is all about.  It would be a limited really site if "Apple Didn't Ship It That Way" was really a major worry.

Technical theories are nice but real world use and specific testing has to be the final derminat.  Those telling others not to beleive there own eyes should know that.

I always say in end to just test the subject of interest and answer the question.  If someone doesn't believe the answer they really need a different occupation or hobby.

I then offered my experience with PCI QE, the one detailed on this site.

I also give the obvious corollary example pointing out the the PM8500 does better using the "technically much more constricted" SCSI drive as the boot drive than using an ATA drive.

A similar "measured" vs "real world use" effect can be seen with the drives.  The ATA 100 and 133 drives connected to the PCI card bench over twice as fast an the large SCSI drive (An ATLAS 10K 68 pin connected the the internal SCSI 2 port.)   However the system runs MUCH better is booted from that SCSI drive.  This is almost certainly related to virtual memory and SCSI more advanced abilities (bi-directional, less CPU load).  If I refused to believe my own observations the NUMBERS would require me to boot from an ATA drive.

I proceed to hint at a theory provided by me, a toal non-computer professional an amateur, a pretender, a fact that proves that old adage (or maybe I just made it up, but it sounds old):  It is better to know how to think than to think you know... (Great, no doubt sometime within the next year I will hear an AA member ascribe that quote to Bill Wilson.  He already got credit for 'We have nothing to fear but fear itself.')

Ignoring the fact that telling people actually using hardware that they are dreaming and fools is rude and condescending and the overall approach is bad science and bad practice lets look at this analytically.

Having read nothing from Apple or anything much about this particular technical issue but instead reallying on the observations and experiences or myselft and others let me toss out some ideas.

The obvious first thing is to ask is: What able DID do to evaluate and decide not to enable PCI to use QE, and why.  Although Apple did NOT enable QE extreme on PCI macs by default, they certainly didn't discourage it.  

Turning on QE on my PM8500 was a heck of alot easier than 1) installing OS X on it, 2) getting CGI
s to work, or 3) setting up webDAV (not too hard).

The point is, enabling QE is easy.

When you further ask yourself what hardware Apple must have been looing to decide if QE was a good idea on PCI Macs you do not suspect they tested is on1st generation PCI PowerMacs, systems which were never supported by ANY shipping OS X.  Even the Public beta only installed as "unsupported".  Instead they probalby consdiders machines that where much newer and supported buy OS X, like the B&W and low end (not sawtooth) G3 and G4 systems.

So now we ask: How did thos systems differs form the original PCI Macs, and the answer is: They differ in a lot of ways the overal balance of which is hard to predict.  A few of these differences are:  Faster buses,  even faster memory, more and different interfaces (like USB, FireWire)g with a trend away form older interfaces like (SCSI and serial), ATA buses, and PCI slots conforming to an upgraded spec.

When we look at my experience, especially when combined the experience of others, we generally DO NOT see the PCI QE enabled video card sucking up all the bandwidth and causing problems in these original systems.  In fact we don't see any PCI card dominating the data flow from the motherboard, drives, are other cards. What we do see is bandwidth problems between interfaces arrising of the same card so that USB devices attached to one card might have problems, or a card with both USB and FireWire might get choppy USB mouse when FireWire drive is attached, etc.

If you look at the results from users who have upgraded the Radeon 70000 32 M on there first generation PCI systems to the newer much faster Radeon 9200 Cards with 128 M of VRAM yu find that generally they are seeing NO performance gains at all.  This is surprising since QE benefits so much on newer systems from having more memory on the Video card.

All this suggesst that these old systems are aready feeding as much data as they possible can to the current cards and that is not enough data to overwhelm the motheroad and tends to point to the the actual bottle neck (or an important one) in a 1st general PCI PowerMac being the connection between the motherboard and the PCI cards.

Does this track...of course it does.

The first generation of PCI Macs had 2.0 compliant PCI slots.  Subsequent generations had newer 2.1 slots.  That difference seems small, but anything but. The 2.1 spsec was a major improvement over 2.0 and 2.1 with buses that can manage and move a lot more data than a 2.0 bus can.  

Think of the different ATA drive benchmarks given by the Beige and the PM8500.  Using identical hardware the PM8500 can access the drive at a max of about 20 MBytes/sec whereas the Beige gets 40 to 50 MBytes per second.  The Beige really got a MAJOR performance boost from an ATA card but only moderate boost from QE.  The PM8500 actually runs worse booting from an ATA drive  (working through a PCI slot) doing bettter from a good SCSI drive even though that drive only benches at arond 9 MBytes per second (near saturation for a SCSI 2 bus).  The PM8500 gets a much more signifance performance gains from using QE than from and ATA drive.

The Power Center that has a much slower memory system that the PM8500 and the Beige has faster memory and faster PCI slots but bopth seem to benefit less from QE than the PM8500 does suggesting (very softly) the the PowerSurge system may be a special group as pertains to QE.

Apple looked at newer PCI system that had faster PCI slots and saw the potential for QE to pull too much bandwidth starving out other devices.  It is not clear if this was a constant problem, or just one ont he margins,  but not one for Mom to contend with.  They probably never looked at, didn't care about, and even if they did would never comment a system as old as the PM8500.

However a first generation PCI Macs hobbled PCI interface is so slow is may actually insulate the rest of the system from any particlar PCI card's bandwidth appetites.  That Radeon 7000 just cannot suck enough datat through the straw of a 2.0 connector to cause major problems. but might really imact a system pulling through a PCI 2.1 hose.  This isolating effect allows these olders system to take advantage of QE extreme (it seems enough  data is flowing to keep the GUI pretty fluid) while avoid the bandwidth problems.  Of course the trade-off is lots of bandwidth problems withing the PCI cards (FW/USB/ATA) and/or between the motherboard and and cards, some dating to when these systems were new (audio stuttering from  any sound source passing through a PCI slot).

I found it perplexing HOW brittle those PCI slots were often having problems at data rates well below what the motherboad could easily provide.   A glaring example is DVD video (or any video).  The PM8500 had no problem playing DVD video if that video was copied to a SCSI drive, but just could not handle that same video when read from any device connected through an PCI slot includiing FireWire, USB, and ATA.  We are not talking much buch data here.  My 1st generation PB G4 had NO trouble playind a DVD from a drive attached via its USB 1.1 ports (about 1 MBytes/second max).  The PM8500's serial ports can hit around 200KBytes/second, its external SCSI around 5 MBytes/second, and  the internal drive around 10 M/s.

The ATA drives connected via a PCI CARD can in fact hit simple read/write speeds of 20 MBytes/second, much more than a DVD needs, suggesting a large component or the problem might relate to trouble managing the bandwidth in and among those PCI slots (one of the documented changes form 2.0 to 2.1). This limitation is highlightyed by many of these systems odd and unique problems including: Audio sources that stutter when sourced from a drive connected via PCI  but adressed in OS 9  by using an extension that actually slowed the drive down,  intractable problems using 2.0 devices in but USB 1.1 works perfectly with problems worsing as the possible data flow to/from the device increases.  Connectin a USB 2.0 drive to these machines just freezes them until you unhook the device.   Yet another eample is the poor function from the the FireWire connected iSight as it overloaded its bus eventaully failing altogether, but leaving the system otherwise functional.  Restoring temporary iSight function always required a reboot.

Clearly some cards worl better than others and in general the problems are more like or more pronounced for data flowing into, rather than out of the system.  Audio and video coming in stutters, the iSight fails when feeding in raw videa as well,  but Video on Demand (using the EXACT SAME FILES) works fine, you can burn to ATA CD drive at 24x, you can output to a printer, and your dump QE Open GL command and textures to the video card as adequate speeds.

It is not enough to know the generalties about some theory or evauation.  If you are going to swear buy it you had better make sure the model for hardware and how it is used matches what you are likely to do.   In other words its all modeling and you need to make sure the model relfects your situation.

Everybody's needs patterns of us and peeves are different making pros and cons somewhat varialbe, so believe yourself.

Theory and reality often don't match.  That is the grouding principal of science: If your theory starts to diverge form your observations you don't try to force your observations to match your theories, and you don't ingore the observations.  You modify you theory.  Over time this process hones your model guding and closer and to the reality.

When in doubt and when really want to know... ask the questoin.  Who cares if you predicted wrong, the fun parts is explaining why...otherwise nobody learns a thing.