OK...I think I have got an idea how this thing ticks....
I tried 
9x300 w/ 133divider for 400ddr cas 2 (stable speeds)
3417
then I tried 
8x325 w/ 133 divider for 2600mhz but running 433ddr speeds cas 2 ( not prime stable)
3316
Note the thing here is that at 2646mhz (8x294) w/ 133 divider for 392ddr cas 2 I only got a 3291....
So here the added memory speed and fsb which probably translated into higher HTT 3x325 = 975 vs 3x294 or 882...both were cas 2,3,3,6 timings
then I tried
8x337 w/ 133 divider for 2700mhz but had to run at cas 2.5 for 448ddr or it wouldn't post...stable to run benches but not superpi or prime
3431
Interestng thing to note here I didn't gain the jump 2600mhz with higher ddr did versus 2646 with lower ram...likely do to cas timings....
So I think 1:1 ratio high ddr system with good clocks will get good scores. Problem is the boost gained for high ddr starts to get a bit lost with lower cas timings it requires...pretty much what we see in real world apps and what I have been saying all along...