1. I didn't say (at least I don't think I said) that you can't have
per-part encodings, merely that you can't encode certain parts, notably
multipart -- that is, in multipart messages you push the encoding down
to the parts. That sounds like it should satisfy your concerns very
nicely, I would think.
2. If all the participants were running either Andrew or my new
MAGICMAIL software, surveys by mail would be much easier to take surveys
by mail. I don't know of any good mechanisms that work with arbitrary
existing user agents, however.