I have a patch for this, which uses MBCS conversion instead of relying
on the default utf-8 (here and several other places). Tested on a French
version of winXP.

Which leads me to the question: should Windows use MBCS encoding by
default when converting between char* and PyUnicode, and not utf-8?
There are some other tracker items which would benefit from this.

After all, C strings can only come from 1) python code, 2) system i/o
and messages, and 3) constants in source code.
IMO, 1) can use the representation it prefers, 2) would clearly lead to
less error if handled as MBCS and 3) only uses 7bit ascii.
There is very little need for utf-8 here.