That's a very interesting question. I think that the spec isn't very
explicit, but the rule about nested encodings gives a good clue: You're
not allowed to have any C-T-E on multipart/* or message/* unless it is
"7bit" "8bit" or "binary" -- and these three are basically synonyms for
"none". (That is, there's no encoding, but the C-T-E header makes a
statement about the assumed underlying transport system.) If you think
of these three as being "none", then the answer to your question is
trivial -- there are no nested or inherited encodings, period, and the
"none" synonyms are just ignored.
Does that make sense? Does it help?