I've attached a multipart message produced by Thunderbird. For some reason, the email.message parser doesn't properly recognize the boundary. This causes legitimate multipart messages to parse as if no boundary were specified. Simply removing the newline on the line starting with " boundary" allows it to parse correctly.

To see why this is valid, refer to section 2.2.3 of RFC2822 [1].

[1] http://tools.ietf.org/html/rfc2822#section-2.2.3