This is mostly speculation, because I lost all my benchmark data involving this, but I just recalled now; the FXAA in the shiny new 300 series of Nvidia drivers actually seems to incur a performance loss rather than a gain in most things i've tested. I got roughly 400 points more in Heaven benchmark without, and a couple of frames more in most synthetic tests I've run when i'm using standard AA or FXAA, dependant on the test. Weirdly, I know a friend with a 670, clocked up pretty damn high, and he's pulled similar results. FXAA actually leads to less performance than traditional MSAA or similar. Whether it's down to Nvidia still working on the FXAA code or games simply refusing to support it correctly; I don't know, but it does seem to be slower in everything I can test. However; that's only SLi'd 550s and a single 670 i've tested with. I'd be grateful if people could test on their own systems, just so I, and a few others, know once and for all if FXAA is actually worth it.
You must be benching wrong. FXAA is incurs roughly 10% hit on FPS as opposed to no AA at all. 4xMSAA will cost you 30%+ FXAA incurs more of a hit when you force it via the driver instead of enabling it in game, but its still certainly no where near the hit you take if you apply MSAA.
with FXAA enabled i get much highest scores in heaven vs 4xMSAA BF3 with 4xMSAA will struggle for 60 fps on a 680 with FXAA enabled your in the 80s
Perhaps it was an issue with the 301.42 drivers we were both testing with. I still fail to understand how a 670 can be slower with FXAA than without. It's partly why I asked people to test for themselves.