That PCPer bench is very odd. In no instance should a 285 ever be faster than a 5870 over a full benchmark. In one case, I even saw it faster than the 480.
Something strange is going on there.
I like to see minimums as well, but something to take into account is that higher averages usually mean higher minimums, also. What you see on a bar graph for a minimum represents a single frame--not so for the average. Going by minimums only can be tricky unless viewing a line graph that shows trending, because that minimum on a bar graph could have lasted for a single frame. And it could've been a hard drive access issue or something totally unrelated to GPU performance.
Any analysis on minimums really needs to be trended by a line graph to show a repeating pattern.
A graph like this is what's needed to view minimums. Here you can clearly see the difference over a period of time between 480 SLI and 5970 on Metro when tesselation is enabled. Seems to be hitting 5970 a lot harder right now.