Thanks for the note David. Zimbra use's the subject and a date range to create our conversations. Your right that this does cause a few improper groupings. There is at least one open issue on that:
http://bugzilla.zimbra.com/show_bug.cgi?id=5935
Technically speaking we'd need to store a bit more data in the database to make conversations based something more that subject and date. At this time we don't have to store *all* the headers in the database, so this would increase the complexity a little.