Purposely leaving out games that are skewed toward Nvidia is actually unfair to Nvidia as there really are a fair amount of games that do so and people do play them so they need to be factored in. I think the games Techpowerup have chosen to use are rather representative of the gaming market as a whole. That said I'm not sure why you are claiming bias actually because of the three the Hardware Canucks review has the game selection that is most skewed towards Nvidia IMO.
The guru3D link you gave was to one game. If you average all the games in that review they have the GTX 460 768mb as 23% faster than the HD5770. I don't feel like doing the math for the HD6850 but I'll give you that their numbers do appear closer to the 15% you are saying than my "less than 10%."
The Hardware Canucks review you are reading entirely wrong I believe. That chart is saying the GTX 460 FTW card which is sold with a huge overclock is 15% faster than the OC HD6850 at 1920 with 4xAA, not slower. OCed numbers are interesting but not really appropriate to what we are discussing. Here is the conclusion of their recent HD6790 review instead because it uses cards all three of the cards we are discussing at reference speeds;
http://www.hardwarecanucks.com/forum/hardware-canucks-reviews/42228-amd-radeon-hd-6790-1gb-review-19.html
It says the HD6970, at 1920 with 4xAA, is 5% slower than the GTX 460 768mb and 15% faster than the HD5770. So that's saying the GTX 460 768mb is 20% faster than the HD5770 basically. It also says the HD6790 is 14% slower than the HD6850 vs 5% for the GTX 460. The difference there is 9%. Straight addition/subtraction of the percentages with different baselines like that is actually mathematically incorrect but it should still be very close to the real numbers, maybe off by a percentage point either way, but in any case their numbers are quite close to what I'm saying.