Rendering performance has been greatly improved on Windows platforms thanks to the
new GPU-accelerated D3D pipeline in 6u10.  However, there still remain many machines
(maybe 50% or more of the Windows market) that are unable to make use of the
D3D pipeline due to insufficient drivers and/or hardware (e.g. Intel integrated
graphics chipsets).  For those machines without D3D support, we use the software-based pipelines by default, so it is essential that we improve performance of the
software-based code paths to bring them closer the performance level of the
GPU-accelerated paths.  Adding MMX and/or SSE optimized versions of our software
loops, similar to what was done with our VIS loops on SPARC platforms,
would go a long way to improving rendering performance of many important cases
on x86/x64 hardware.