21 & 22 Cinebench R10 released in 2006 is next. It supports up to 8 threads so dual cores are getting a good workout! Scaling is not 100% but rather in the high 80's! Both single core and multi-core scores are taking into account.
A. NFORCE4
B. K8T890

It seems that scaling is slightly better on the VIA board by 3%! Within the margin of error but nonetheless this is the second victory for AX8 after the one in the sound tests of 3dmark 2003!

23. Wprime 2.10 is a new addition to the test suite. It maxes out the 2 cores on the 4800. It works just fine on XP, meanwhile I could not get it to run on Windows 8.1. A shorter time constitutes a better result!
A. NFORCE4
B. K8T890

Both nforce4 and K8T890 are for all intents and purposes equals as they should be!

24. 7Zip 9.20 is another new addition! Also dual core aware, it scales in the low 80's.
A. NFORCE4
B. K8T890

A higher score is better! Again AX8 has a very small advantage over AN8: 2.5% is within the margin of error. There are, as of now, no instances where VIA manged to win decisively against NVIDIA!

I run wPrime on Win10 x64, it just needs to be "run as Admin" from context menu to work properly.

Alternative programs that can be run under XP : Cinebench 2003, R11.5 (should work under XP), SuperPI 1M + 32M (1M tests L2 more, 32M test IMC/RAM speed more).

Thank you sir, that did it! I ran Wprime with admin rights and it's OK.
Cinebench 2003 - used where I was testing socket A, socket 754 and socket 939 single cores CPU's. It will be back soon enough. I never used Cinebench R11.5 - it should be very tough on these early dual cores, already R10 brings both platforms to their knees.
I used Super pi 1m for single cores but with dual cores i considered Wprime better as it dual core aware.

25 & 26 & 27 & 28. PCMARK 2004 is next. I eliminated the 2002 version as it does not know how to use dual cores while the 2004 one knows how to use dual cores in certain tests.
A. NFORCE4
B. K8T890

WHAT?? CPU and memory scores are very close as they should be however NFORCE4 wipes the floor with VIA in the 3d tests! That means a single thing: pci-express implementation in nforce4 is far superior to the one found in K8T890.
How comes the difference was so small back in the single core benchmarks? I was using a radeon X1950 pro, a vastly inferior graphic card that failed to show the difference in pci-express implementation.
That means that in the benchmarks that tests 3d performance - and most of them test exactly that - nforce4 has an advantage over the competition. This is the explanation for the better scores in 3dmarks!
Meanwhile in the storage tests the tables turn and VIA comes on top with a 7% lead just like in the socket 754 benchmarks where NV8 got defeated by KV8 PRO in both version of PCMARK.

You could check PCIe bandwidth with GPGPU test under AIDA64/Everest Ultimate, however it probably won't work under WinXP.