Well I know the sandra score from me at 2.6ghz from 3:2 ratio to 5:4 ratio is 1000 points or about 16%...Fortunately real world is not sandra memory score....
I think from 3:2 to 1:1 could be 5%, and I say could cause cas latency also has a big effect in the real world. What I mean is yes I run 3:2 now cause at 289fsb my ram can't do it. However I can run 2,3,3,8 with 1t and 2.7v of vdimm. For me to get 5:4 (which i can run but not stable) I need ram to run cas 2.5 with super low settings and 2.85v of vdimm....difference in superpi at 1mb was 1 sec or about 3%. BUt that wasn't stable....I need ram sticks cable of cas3 (the geils just wont boot at cas 3 even at stock speeds). With cas 3 I would take another hit and likely may have cancelled out the 3% gain.
With this being said 400ddr with cas 2,2,6 which some can do versus a 1:1 ratio at 2.6-2.7ghz which needs 600ddr is going to need cas3 settings and a load of vdimm....In the end maybe5% in memory intensive apps, less in pretty round apps, and maybe even a loss in performance in very few apps.
The story for me is it worth it to:
a) buy more expensive ram to run 480ddr and have neglible 3% gain, or evenmore expensive ram to run cas 3 600ddr for possibly a 5% gain?
b) reduce my clock spee (fsb) down to get to a point where I can get the geil sticks to run 5:4....
With A I think the answer is NO...slower ram but at super tight timings is good enough is pretty much all apps. The A64 is pretty bandwidth hefty already...
With B the answer is also NO....higher clock speed versus higher ram speed is a myth that has been perpetuated by many but in test outside of sandra in 3dmark doesn't hold water. Unless I could get it stable within 50mhz or 5-6fsb it wont beat the slower ram with tighter timings to run ram speed about 80mhz faster but more conservative cas 2.5.....Doesn't pan out....