I suspect your CPU limited, because Q3 uses only one CPU, you can force it by console command "r_smp 1", but there's a big chance that the system will freeze or crash after that when trying to run again Q3. But I think you can try it.

One more question, you sau you run 1024x768 ? 32 bit ? All maxed out (texture, geometry) ? Because if that's the case, don't expect too many fps in Q3, it's not a 3dfx friendly game, like UT.

I did more Quake 3 bench marking test with my V5 5500 AGP.  I didn't do a a super detailed thorough benchmarking test, just a quick look comparison.

I'm running the 5500 in a system that has a 2.8 ghz P4, Soyo P4S Dragon (SIS 645D chipset), 512 MB DDR SDRAM, and Windows Me.  I shouldn't have to worry about any CPU bottlenecks.  I also set it to use single chip rendering in 3Dfx tools.  I used medium settings for both geometry and textures.  Also did 32 bit rendering.  Here's what I found:

V5 5500 (single chip)
640x480 - 95 fps
800x600 - 63 fps
1024x768 - 39 fps
1280x1024 - 22 fps
(note: enabling "fasted performance" i.e. dual chips with no AA almost doubled these numbers on this test system)

Here are the numbers again for the 8232 w/dual P3 1 Ghz, Tyan S2567 mobo, and 256 mb PC-133 ECC SDRAM.  Enabling dual CPUs causes the system to hang, so no joy there.  I also ran it at 1280x1024 just to push it a little harder.

AA5 8232 (0 or 8X AA..makes no diff)
640x480 - 77 fps
800x600 - 56 fps
1024x768 - 34 fps
1280x1024 - 19 fps

My preliminary conclusion is the 8232 seems to perform much like a V4 4500 in terms of fill rate (This is a far cry from the fill rates quoted in the owner's manual).  However, I didn't actually dig out my V4 4500 to test that, but I'm sure the numbers will be similar to the non-SLI V5 numbers I quoted above. 

The small differences between the two could be because the single chip V5 is on the AGP bus and the 8232 is on PCI (which has been seen with  V4/V5 AGP vs PCI benchmarks).

I did my research beforehand, so I didn't expect twice the performance of a V5 6000 or anything like that.  There's very little info out on the internet concerning these Q3D systems.  I figured it would have at least been like a 5500 with no performance hit for AA.  The sad thing is my all of my 12 MB V2 SLI pairs will beat the 8232 by ~20% in raw performance running Quake 3.  This leads me to believe that some of Quantum3D's earlier products probably outperform the 8232 as well.  This is somewhat shocking considering the hefty price these units were fetching at the time (my card has a week 36, 2002 date on it). 

It still makes me wonder if there are any registry hacks that can modify the performance of this card.  It could also be related to the fact I had to cheat to get it to run openGL games in the first place.  I had to rename 3dfxOGL.dll to opengl32.dll and drop it in the Quake 3 folder.  I did try the wickedGL mini GL and it was slightly faster, but the image quality was very poor.  I also tried to rename the Q3D OpenGVS dll (as described in the owner's manual), but I didn't have much success there (Quake 3 crashed and Quake 2 sorta ran, but was very prone to crashing). 

Ken