I don't think a new type is needed. I believe that using
application/octet-stream will work fine. The CRLF issue may be a
problem in some contexts if you use base64 for transfer-encoding, but
should NEVER be a problem if you use quoted-printable. (That's what the
recent debate about newline-in-quoted-printable was all about.)
So I guess my recommendation for the case you describe would be
application/octet-stream with the q-p encoding. A UA can do something
quite intelligent with this. In the second metamail release, for
example, users will be given the choice of writing such pieces out as
files OR viewing them as text on their screen, which is probably the
kind of choice that is best left in the hands of the user anyway.