Daniel: i can't remember the exact scenario (i filled this bug 2 years
ago !)
after having a look back at email.message.Message.get_payload, i
remember the problem: the decoding errors are silented by the method and
you have no way to know if the decoding has been successful or not.
to find my malformed email i had to patch the module like that
(basically just removing the try/except blocks):

<patch>
--- message.py.orig	2009-04-03 23:46:47.000000000 +0200
+++ message.py	2009-04-03 23:48:27.000000000 +0200
@@ -192,19 +192,11 @@
             if cte == 'quoted-printable':
                 return utils._qdecode(payload)
             elif cte == 'base64':
-                try:
-                    return utils._bdecode(payload)
-                except binascii.Error:
-                    # Incorrect padding
-                    return payload
+                return utils._bdecode(payload)
             elif cte in ('x-uuencode', 'uuencode', 'uue', 'x-uue'):
                 sfp = StringIO()
-                try:
-                    uu.decode(StringIO(payload+'\n'), sfp, quiet=True)
-                    payload = sfp.getvalue()
-                except uu.Error:
-                    # Some decoding problem
-                    return payload
+                uu.decode(StringIO(payload+'\n'), sfp, quiet=True)
+                return sfp.getvalue()
         # Everything else, including encodings with 8bit or 7bit are
returned
         # unchanged.
         return payload
</patch>

once again, the behaviour is documented, so it's not really a bug.
but it caused me a lot of trouble (and it does not conforms very well to
"Errors should never pass silently.":)

but i guess applying such a patch could potentially break number of
client code so... is it worth the change?