The attached patch is a possible fix; it uses the decode_header() and make_header() functions to figure out the encoding properly; it fixes my example, at least.  But does it increase the odds of crashing on messages with malformed headers?  Should it go into 2.7 given that we're at the RC stage?  What about 2.6?

(BTW, Barry, I noticed this because messages being sent through Mailman were coming out with broken subject lines.  The system generating the messages is slightly weird -- doing the UTF-8 quoting is unnecessary since the subject contains no special characters -- but I think Mailman shouldn't be breaking subject lines.  I haven't verified that this Python fix actually fixes Mailman, but I think this is a Python bug, not a Mailman bug.)