As for benchmarks. With 225fsb for the cpu, VDDR=2.1V i get the following:
DDR 450 (1:1), 3-3-3-10-4, everest gives 5900(read), 2243(write), 79.6 ns
DDR 600 (4:3), 3-3-3-10-4, everest gives 6487(read), 2575(write), 74.9 ns
DDR 675 (3:2), 4-4-4-12-4, everest gives 6595(read), 2656(write), 74.4 ns
DDR 675 (3:2), 5-5-5-12-5, everest gives 6496(read), 2629(write), 78.1 ns

So any divider gives better results than 1:1. And DDR675 (3:2) with very high memory timings gives a better latency than DDR450 at low memory timings. Even with 5-5-5-12-5 the 3:2 divider gives a lower latency than 1:1 at 3-3-3-10-4. So i do not understand why i should avoid the 4:3, or the 3:2 divider?