it just seems that DDR increases raise performance quite linearly with increases in FSB.
To assess performance, all one needs to do is add the DDR MHz with the clock speed in MHz.
Combinations with the highest total will provide best performance. This seems obvious but what is interesting is that there seems to be an incredible number of possible combinations for each performance 'standard'. With more cpu/dram ratios, we could discover all these combinations. Someday, maybe someone will make boards capable of handling 1 MHz adjustments to RAM.
eg.
2720 MHz and 452 MHz DDR translates to 3172 MHz total
this should perform similarly to a:
2880 MHz and 292 MHz DDR
2772 MHz and 400 MHz DDR
etc.
below, we can also see that:
2560mhz 160fsb 3:4 426DDR total MHz: 2986
2655mhz 166fsb 1:1 332DDR total MHz: 2987
which adds up since they have similar real world results.
just my two cents. i obviously had a little too much fun with thugsrook's results
edit: now of course, this would only be true with a particular setup, since no two chips of RAM are equivalent.
That is, RAM chip 1 may hit DDR400 at fast settings while RAM chip 2 can only hit DDR333 at fast settings.
So total MHz wouldn't be a good way to assess performance between machines. Simply a quick way of working out the best overclocking parameters.