I think Icetech's results demonstrate that a card's average performance doesn't necessarily predict what it will do in particular circumstances. On average, Silencer is right that an HD4850 is only a little better than an 8800GTS. But in this particular situation, for some reason, it's apparently a lot better.

You can see this sort of thing pretty frequently on charts in hardware reviews. They'll test a card with six or eight games and on one of the games, the card performs much better or worse than it does on average.

Here are some nice charts of average performance for a lot of cards (although they don't include the 8800GT or 8800GTS):

http://www.tomshardware.com/reviews/...0,1964-19.html