An overclocked GTX470 does scale well, but please don't spin the numbers the way you want to see them. Here are actual results:
1. 5870 was faster than an OCed GTX470 in 3 out of 8 games: BF:BC2 (53 vs. 52 fps = +2%), Dirt 2 (76 vs. 69 fps = +10%), and Anno: 1404 (75 vs. 66 fps = +14%). So that means a GTX470 OC at most won 5 out of 8 games,
or 63%. How did you get 90%?
2. Examining the games in which GTX470 OCed was faster, the performance difference was less than when 5870 won and hardly noticeable to any gamer. These included: CoD : MW2 (121 fp vs. 127 = +5%), RE5 (109 vs. 114 fps = +5%), and Crysis: Warhead (45 vs. 48 fp = +7%).
So more or less in 6 out of 8 games tested, the performance between 5870 and a 27% overclocked GTX470 was rather similar, with GTX470 OCed never exceeding the 5870 by more than 5%, and 5870 never exceeding GTX470 OCed by more than 14%.
3. This means a 27% overclocked GTX470 only convingly provided better framerate in Far Cry 2 (+32%) and Just Cause 2 (+22%). This is not surprising of course because even a GTX285 is faster than 5850 in both of those games. Of course if the reviewer chose STALKER: CoP:
http://www.xbitlabs.com/articles/video/display/geforce-gtx-480_9.html#sect1 or Stormrise:
http://www.xbitlabs.com/articles/video/display/geforce-gtx-480_7.html#sect3, then the conclusion would have been reversed.
The point is some games run faster on NV and some games run faster on ATI. Choose a videocard which performs best in the games you want to play.
Having said that, it is hard to recommend 5850s at anything above $320+ considering how poorly 5xxx series performs in any title with tessallation (Metro 2033, Alien vs. Predator, etc.). At $280-$290, 5850 is easier to recommend vs. a $350 GTX470.