I'm not so sure the situation is quite as bad as they make it out to be.
Draw calls / state changes were heavy in DX9, yup, but considerably cheaper in OGL and since DX10. The numbers they cite for consoles is far from an impossibility on PC... not that it's intelligent to just completely ignore the possibility of batching render states and shit to improve performance.
I would think if your engine is intelligently designed you will run into other problems long before API overhead ever becomes a problem. If ATI / NV ever get their driver teams in gear then this stuff will only become cheaper as well. The threaded draw call submission is not functional in DX11, there's no driver support yet from either parties.
Draw calls / state changes were heavy in DX9, yup, but considerably cheaper in OGL and since DX10. The numbers they cite for consoles is far from an impossibility on PC... not that it's intelligent to just completely ignore the possibility of batching render states and shit to improve performance.
I would think if your engine is intelligently designed you will run into other problems long before API overhead ever becomes a problem. If ATI / NV ever get their driver teams in gear then this stuff will only become cheaper as well. The threaded draw call submission is not functional in DX11, there's no driver support yet from either parties.