It sounds like rmail.el currently doesn't bother to look at the
Charset= header before it decodes a message in Babyl format; it only
does that for mbox-format messages. This backfires when one fetches
email via POP with movemail, since movemail writes Babyl-format
messages in that case. Quoted-printable decoding is also not done for
Babyl-format messages, so movemail/POP loses again.
In addition, if the message is encoded in base64, rmail doesn't decode
it as it does with quoted-printable. This loses with some wacky mail
servers (such as Exchange) that tend to encode messages with non-ASCII
characters in base64, even if the message has only a single part and
uses 8-bit charsets perfectly suitable for quoted-printable.
Would it be a good idea to fix these deficiencies? I have a patch to
rmail.el that fixes all of them (I'm using the patched rmail for the
last couple of months). If the idea sounds reasonable, I will install
the changes in the CVS.