That's an interesting link. They forgot to mention the socket used, but that can be figured out... the 3000+ is a Winchester, but close enough. The 3700+ wins on almost everything. Notable exceptions are any benchmark with the word "memory" in it, four benchmarks labeled "Wstream" something or other and "3DMark05 Graphics." My thought on the last item is this: 3DMark05 tried to test graphics card only. Thus, the scores for various CPUs are almost identical, however the socket 939 does beat socket 754 by a whole 8 points. Why? Well, I'm thinking since RAM is faster in socket 939, 3DMark05 may use more textures than fit into the frame buffer, thus using system RAM. In this case, if system RAM is faster, faster graphics performance.
In a real-world game where the CPU is used to do stuff like computer AI control and such, the higher GHz CPU wins - in this case the socket 754 3700+. Then again, why does the 3400+ Newcastle with 512K cache beat the 1MB cache 3700+ Clawhammer in 3DMark05 with 5 points when they both used the same motherboard and video card? Strangeness...
Checking some other numbers... The socket 754 3400+ beats the socket 939 3200+ (as it should according to AMD's rating) in much the same way, winning all the benchmarks except for a near loss in 3DMark05 and losses in the "memory" and "Wstream." These two CPUs are around the same price.
The socket 939 3500+ and socket 754 3400+ was a closer battle. Again, everything "memory" went to socket 939. With this comparison the slower GHz socket 939 manages to take some more wins away from socket 754. It very well should, since AMD rates it higher. AMD also prices it higher. Still, the MUCH cheaper socket 754 CPU takes over half the benchmarks. The cost difference between the two can buy a Sempron 2500+ as a spare CPU. :laugh:
Overclocking/upgrades aside