It's not about whose fault is it, it's about end user results. We need benchmarks that use the default graphics stack for a platform/game. If a game targets DirectX primarily and is optimized for it then it should be DirectX that is used for the benchmark, not the unoptimized code that targets OpenGL. End users don't care whether DirectX or OpenGL is used, they only care about end results.
Originally Posted by ua=42
Speaking of end user results, this guy made a quick comparison for Metro:LL between Windows and SteamOS and it seems it's performing pretty much equally on both platforms, which is great to see (scroll down to the last video):