Not meaning to stir the pot or anything, but I believe that 'section' for testing as well as the benchmark methodology just happens to run noticeably lower fps on AMD cards for a light combat zone rather than being relative to the majority of combat focused gameplay where AMD seems to run much better(at least relative to the above averages).
I took screenshots of the same exploding barrel from a slightly different angle for example(so that mech in the background isnt in full view) and the result is significantly different, same settings.
Just sharing so users can see for themselves it's not stuff like barrels causing the AMD cards to drop fps and actually combat fps(while it does have drops here and there) is much stronger in most other areas of the first level, I actually benched each combat zone of the 1st level,, though I did not use 3 round passes for all areas, I did do multiple runs for some of them and still came out with some level of consistency even with varying lengths of time in the same zone.
I focused on 'combat zones' that have to be cleared to progress and only benched from beginning to end of combat to avoid staring at empty scenery, though obviously my method was a little different, it was definitely much more action packed testing that might be valuable for some.
The theory that nvidia may have optimized something to do with background scenery in early stages or maybe even further throughout the game seems pretty strong after further testing, though I understand & respect Steve's method to try and eliminate 'randomness' that we get with AI for consistency.
At the same time I feel when the scenery is the focus of the benchmark, the results might favor Nvidia,, a more action-oriented approach to benching might produce much more competitive results from AMD but I understand time constraints & consistency may be a little harder to nail down, I'm also not saying it
HAS to be done as the existing benchmarks are still perfectly valid
for the method used,,, simply sharing my 2 cents in-case the information can be used(or investigated) to improve future testing when comparing Nvidia & AMD performance.
'Driver trickery' to favor one form of bench-marking over another is a strong possibility imo. I'm actually curious if nvidia GPUs performance margin remains the same in heavier combat. Cheers.