Well, a caution that tweaking the regex can have unexpected consequences as past issues have proven (but by all means go for it), and a note that the parsing strategy is going to change completely in email6 (see http://pypi.python.org/email and http://hg.python.org/features/email6).  I think your tests should pass on that branch; I'll be interested to try it when I get some time.

(Note: I'm removing 3.1 from versions since it doesn't get bug fixes any more.)

Also, I'm not sure the (non-essential) change to consolidate like-charset encoded words is appropriate for a bug fix.  It's hard to see how it would break anything, but why take the risk if it isn't needed to fix the bug.