40 & 41 & 42. Doom 3 built on IDTECH4 engine is the next game to be tested. Unlike previous games it is extremely tough on the video card; no period correct GPU can ever hope to play it on max details; in fact no DirectX9 GPU could put up 60fps using maximum settings, not even at 640*480. It is single threaded only so using an Athlon 64 X2 does not deliver superior performance to an Athlon 64 4000.
It is the first game I know that allows AA to be set at 16X! Because we would be GPU-limited at such high levels of aliasing I limited the maximum AA level to 8X in all synthetic benchmarks and games. That means that we are CPU-limited - so GTX 285 can not fully stretch its legs at such "low level" of anti-aliasing.
A good pci-express implementation is mandatory of you want high frame-rates in Doom 3 at these settings with such heavy hitters as GTX 285- by now it should be clear that i am writing all this because i am going again to contradict all established knowledge!

For testing I used the HardwareOC benchmark tool for the first time! Here we go:
A. NFORCE4
B. K8T890

And now everyone: WHAAAT??? How is that even possible?
NVIDIA wipes the floor with VIA in Doom 3! Third game in and nforce4 managed to score already decisively in 2 games! nforce4' s performance until now has been outstanding! The only point of contention is the poor GUI performance in Tom2D.

I wonder why nforce4 system used just 2 GB RAM, doesn't it enable CR 1T ? VIA has 4GB RAM in Doom3. Also memory timings are different.

That is because I removed the Corsair XMS pro in order to install it in the AX8 and installed some "normal" Corsair modules. Originally I saved the HTML page with the results but forgot to capture the screenshot with irfan view. After I managed to acquire another GTX 285 - it seems people do not realize how useful is this card with late Windows Xp era games and always want some geforce 7900 something - I re-pasted it and let it run 3dmark 2006 overnight to check the temperature and stability.
Retested this morning in original configuration again before heading up for work:
Abit AN8 SLI +Athlon 64 X2 4800 + Corsair xms pro + gtx 285

Second GTX 285 in working order after some TLC; it seems they all follow reference design and come with the same GPU/shader/memory clocks:

Please note the enormity of the situation: with 8X AA GTX 285 is actually limited by the Athlon 64 X2 4800. However the situation reverses with 16AA: Doom 3 - a game from 2004 - is limited by the GTX 285 - a card from 2009! It seems that anti-aliasing is the Achilles heel even for the most powerful DirectX10 GPU ever released!
If you disable msi afterburner and FRAPS you get an average of 60 FPS at 1600*1200, you will get some dips in smoke/fire/hell's haze filled areas but otherwise you can play Doom 3 at maximum details with a GTX 285.
Like FARCRY the Hardware OC benchmark tool tests the game without sound.

43 & 44 & 45. Splinter Cell Chaos theory is the first game from 2005 to be tested. Released in march 2005, it is very easy on both the CPU and the GPU. The game does not let you select both AA and shader model 3.0; you have to use shader model 1.1 in order to enable AA. Because shaders produce a higher fidelity image than the correction done by AA I tested with shader model 3.0 and AA disabled.

I used the HardwareOC Splinter Cell Chaos Theory Benchmarking Tool version 1.1 and the lighthouse demo. We are completely CPU-limited.
A. NFORCE4
B. K8T890

For all intents and purposes it is practically a tie! There is no advantage for either platform here.

46 & 47 & 48. Serious Sam 2 is next. It is built using Serious Engine 2, a revolutionary game engine back in 2005 that was used only in this particular game! The game supports both HDR and AA at the same time if you are using a GTX 200 series card; from what I remember this is not possible on geforce 6 series, but i might be wrong on this one.
The game utilizes 2 cores at most; one at 100 % and the second at ~50%. I used HardwareOC Serious Sam 2 benchmark tool with the first demo - Branchester; it does not use sound:
Please note that the info on the VIA platform shows a GTS 250 as the card used because i installed this card first and tested with it, however the GPU/memory frequency clearly belong to GTX 285.
A. NFORCE4
B. K8T890

Again we got no clear winner, the small 2-3% advantage for the VIA system is not enough to distance itself from the NVIDIA system. We are CPU-limited again!

49 & 50 & 51. Quake 4 is the second game built on the IDTECH4 engine. A heavily modified and improved version of the engine from Doom 3 it is exceptionally tough on the video card and, in this case, on the Athlon 64. When you are actually playing the game it fully utilizes both cores of the Athlon 64 X2 4800. That means in msi afterburner you will see both cores at 70-90% and also the video card at 100%. It is the second game after Serious Sam 2 to understand and use CPU's with 2 cores. We are CPU and GPU limited!
I used HardwareOC benchmark tool; benchmark does not use sound; all settings including the advanced ones at maximum; AA = 8X:
A. nforce4
B. K8T890

Nforce4 is off to a good start at 1024*768: 6.25% better than VIA is very good actually so the platform choice matters. However we see that despite the increase in resolution from 1280*1024 to 1600*1200 - 46% more pixels - the average FPS drops very little and furthermore NVIDIA and VIA are neck in neck!
That means we are CPU limited! It would be interesting to see what the results are with an Athlon 64 X2 6000!
Quake 4 is the first game where socket 939 feels on the edge! A 2005 game that actually needs much more than the best of the best from the same year to be able to play at max details.

Quake 4, despite supporting dual cores and utilizing them doesn't seem to outperform single core at the same clock running with CR 1T and DDR 533. I got the same score at 1280x1024 and 1600x1200 with GTX 285 and 8x AA. CR 2T in s939 is probably contributing negatively. I think s939 would distance itself in newer games from 2007-2008 with better engines.

I would recommend people with s939 get stocked with GTX 285s in case they die. I got a few GTX 275/260s for s754.

It certainly outperforms any single core Athlon 64! Take into account that msi afterburner 4.5.0, fraps 3.5.99 are running at all times. These consume more CPU cycles than the older versions that I used until now so performance drops by 7-9%.
Also, as you adequately put it, CR2T is dragging the performance down!

52 & 53 & 54. FEAR is the next game, also from 2005. When it was released there was no video card that could drive it to the max! The killer setting is soft shadows; with GTX 285 this is not a problem; max AA settings is 4X, again super easy for the video card. It is single threaded only and we are CPU-limited, again!
I used the builtin benchmark that tests with sound enabled.
1024*768 max details including AA 4x and SS on:
A. nforce4
B. K8T890

1280*1024:
A. nforce4
B. K8T890

1600*1200:
A. nforce4
B. K8T890

There is nothing to see here, moving right along to...

55 & 56 & 57. Call of duty 2 is the last game from 2005. If you applied the latest patch - 1.3 - it will support dual cores or hyper-threaded CPU's and will use the first core 100% and the second one 60-70%. It is the third game that know how to use multi-threaded CPU's - and we are still in 2005! It is extremely CPU-bound; I used HardwareOC benchmark tool with the only test available -it uses sound during benchmarking; all settings to the max:
A. nforce4
B. K8T890

There is nothing to differentiate between either nforce4 or K8T890! Note that during gameplay the frame-rate will actually be much higher because the HardwareOC tool tests at 3X-5X speed from what I see on the screen.