There is another issue with testing on Vista, which I have noticed but 
I don't think apply to this issue:
os.unlink(testfilename)
followed by
open(testfilename, 'w')
often fails, because the unlink() operation on vista doesn't 
immediately reflect on the disk!  You get strange and unpredictable 
regression failures because of this.  It is enough to catch the error, 
sleep for a millisecond and try again to overcome this.  If we start 
regular regression testing on Vista, this is onething that needs 
fixing, probably with a tmpfile utility function in test_support.py

Perhaps CCP could contribute a buildbot with Server2008 (=Vista) to 
test the main branches...  not making any promises.