A very basic calculation has come to my mind... As you know, Cypress was absolutely RAPING GF100 in Performance/mm2. Which should mean that Nvidia should have been able to increase Perf/mm2 much more than AMD did, since they started from horrible levels.

However, when you do the math...

HD5870 -> HD6970: 320mm2 to 390mm2 = 22% die size increase, 40% performance increase => 15% increase in Performance/mm2
GTX 480 -> GTX 580: Same die size = 0% die size increase, 20% performance increase => 20% increase in Performance/mm2

AMD has been able to keep up with the ratio of efficiency increase of Nvidia... and keeping in mind they have started from a much much efficient level already, this is quite a feat. Actually they had a couple of months more time to do this than Nvidia did, yeah, but still.