Confused about going beyond 1:1

Hey all,

First off I realize that what I'm asking about will produce essentially negligible performance gains but I'm bored and interested to learn more about the matter.

So on to the question:
What benefits does overclocking my RAM beyond a 1:1 ratio produce (say 5:6 for example)? Correct me if my thinking is flawed here but how would it increase performance if the FSB can't keep up with the RAM? Wouldn't it just get bottle necked?

For my specific system I'm using an E8500 @ 400 x9.5 (3.8) and I have DDR2 1066 modules but I'm just running them at 800Mhz with tighter timings (4-4-4-12) so that I remain in sync with the FSB. I can and have pushed the CPU above 4Ghz which would take more advantage of the RAM's capabilities but I've settled at 3.8 because it runs stable when I undervolt it to ~1.136v.

Yes and, as said, I have pushed it higher and it works perfectly fine. I'm just curious as to the benefits gained by raising the RAM's frequency (without raising the FSB) so that I go beyond the 1:1 ratio.

For example with my current FSB, I could set the ratio to 5:6 and have a FSB of 400Mhz and set the RAM to 480Mhz.

I wondered that exact thing myself for a while, finally just ballsed up and did some pretty extensive testing with Sandra, DDR2 667-1000 and DDR3 1333-1800 on both C2Q and C2D, though I was using nforce chipsets (750i and 790i) so I can't guarantee this will hold up for you, my findings were that overclocking the memory at SET ratios (preprogrammed) did continue to increase the tested memory bandwidth even past the FSB (though I would be hard pressed to explain how this is even possible). The increases are not large, and in the end I chose 800/CL4 over 1000/CL5 because the overall performance was better (lat 60ns, bnd 6.5 GB/s compared to lat 70ns, bnd 7.0GB/s, but lower CPU arithmetic) On the 790i I ended up pushing all the way to 1800 and sticking there because my 9550 has a much lower multi. However while running the e6700 on that chipset @ 3.5, DDR3 1400 performed much better than 1600 (lat 70ns, bnd 6.5 GB/s compared to lat 90ns, bnd 5.5GB/s) which lead me to believe there was definitely idiosyncracies involved in exceeding the FSB. Also when in unlinked mode on either chipset, even at identical frequencies, the performance took a substantial hit (roughly .5 GB/s) compared to linked modes. Honestly I lean towards the fact that all of our timings are exaggerated (quad pumped, DDR3, etc...) as explaining OCing past the bus, but I could be completely wrong.