Tom's Two Cent
After having a close look at the benchmarks you'll see that the FIC PA-2012 is not the most interesting candidate at all anymore. It performed worst in pretty much every benchmark, although the distance to the other boards isn't big in all except one case. I still wonder what the board has against the 3Dfx Voodoo. I ran this test 5 times, but it kept being almost 10% slower than all the other boards.
Soyo's SY-5ED5 shares one problem with the PA-2012, it doesn't supply the correct voltage when setting it to 2.1 V. I measured 2.44 V and this is simply too much for the new K6 266. This does not have to be a serious issue for long, because I would expect this problem top be solved pretty fast. However it shouldn't have happened in the first place and how do you know if you've got a board with the correct voltage or if you haven't? The performance of the SY-5ED5 is hardly much worse than the performance of the others, but it still couldn't make more than 4th best in the benchmarks.
The surprise candidate to me was the EpoX P55-VP3 . It was the only board that ran straight away, without any fiddeling around. The performance of the board is fine and I am sure that I only couldn't overclock it to 300 MHz because the voltage was with 2.1 V a little bit too low for overclocking. As soon as EpoX could fix the voltage and raise it to 2.2 V I'm sure this will be an easy to use and reliable Socket 7 AGP board.
EpoX P55-VP3: Very easy to configure, but no 83 MHz system bus clock.
DFI's P5XV3 is the fourth candidate with VIA's Apollo VP3 chipset, but as well as FIC's PA-2012 it offers 1 MB of L2 cache and contrary to the PA-2012 it's able to make use of it. The 1 MB L2 cache is most likely the reason why it won almost all benchmarks. Although the manual wouldn't mention a word about 2.0, 2.1 or 2.2 V CPU voltage I eventually could find the according jumper setting for the K6 266. The multiplier setting for the K6 266 is also different to the description in the manual ... if you put the jumpers as said there you'll run the K6 at 333 MHz instead. Nevertheless, this board is currently the fastest Socket 7 board I've tested and if you can live with the fact that it doesn't even support 75 MHz (no good for 6x86MX PR233 owners), let alone 83 MHz, then this is currently the Socket 7 AGP board to go for.
Elitegroup's P5SD-B is currently the only board that came with a different chipset, the 5591 from SiS. I haven't been a big fan of SiS chipsets in the past, because they normally had quite a lot of trouble with the RAM timing. However, this board seems to put an end to this rule. When I received this Baby AT board I couldn't wait testing it, because even on the board it shows the jumper settings for 83/33, 90/30 and 100/33 MHz system/PCI clock. Unfortunately 75 MHz was all that ran reliably, but that could get fixed with a BIOS update. After all this is the first official 100 MHz board I've seen. The performance of the P5SD-B isn't quite as good as the performance of the DFI board, but the P5SD-B only comes with 512kB L2 cache, and for this amount of cache the second place in performance is quite considerable. Of all the five boards I've tested it would be the one I would go for, although it may be smart to wait until 83, 90 and 100 MHz bus speed will run reliably.
Elitegroup P5SD-B: If this doesn't look promising ..