Well... as discussed here, I/we assume that Linux is a lot more efficient at crunching HCC, if the work units for Linux and Windows clients are indeed worth the same amount of research.
HCC clearly is my passion, along with HFCC - so if I had even a single dedicated machine left, I'd put it on Linux and HCC only.
Unfortunately, I've downsized a little over the past half year, and every single one of my machines has a user now who wants and/or needs Windows

But if I were to receive a sudden influx of cash (meaning I'd finally re-start building my own little crunching cluster ), those machines would all do HCC on Linux 64.