According to RFC5987 (http://tools.ietf.org/html/rfc5987), it's possible to use other encoding than ASCII in header fields. 
Specifically in the CGI library, posting files with non-ASCII characters will lead the header to be (for example) filename*=utf-8"xxxxx" which is not recognised:

l 513

if 'filename' in pdict:
    self.filename = pdict['filename']
self._binary_file = self.filename is not None 

The file will thus be treated as a string.
The correction isn't too big but being a total newbie, I'm a bit scared to suggest a patch.