BF3 at 1920x1200, High Quality - GTX580
BF3 at four different resolutions, Ultra Quality - Radeon 7970
BF3 at 1920x1080, Ultra Quality - Radeon 7950
Three different sites, published between Oct 2011 and March 2012. Conclusion? A Core i3 2100 and a Core i5 2500 would be inseparable in benchmarks, playing at IQ settings you would want. Having said that, there are two key things the benchmarks don't reveal:
1) They are all single-player benches. It's often argued that multiplayer 64-player maps are the best test of differences in CPU processing power. I can see the logic in that, but I have no way to test it, and looks like most websites don't either (couldn't find a multiplayer CPU-scaling review with a quick search). I could probably do a few benches on the always-full Operation Metro 64-player server I normally play on (I think that map has the most predictable gameplay elements) at different CPU overclock settings (like, from 2GHz to 4GHz) on my i5 750 but...time.
Maybe someone has done this? On the same (full) server, in the same areas (like the 'B' chokepoint in Operation Metro, with the predictable 64 grenades-exploding-every-second shenanigans? Other maps may have too much variety between indoor/outdoor areas, vehicles and explosions.
2) I have started to see value in how TheTechReport now focuses on 99th percentile frametimes (rather than number of frames rendered per second) and measuring delay spikes rather than the average fps. So maybe yes, a faster CPU may reduce frame latencies even when most of us think we are GPU-limited?
Even if that were the case, I'd still argue you wouldn't do any better than a 2500K @ 4.5 GHz.