Someone who claimed to be Mark M. said on Sat, 1 Jun 1996: (I said:)
Why not simply use two session keys, and encrypt the headers with one while encrypting the actual data with the other? That seems to solve both problems, except that more CPU cycles are required.
An easier solution would be to just strip of the headers. If the header is always the same, then it is redundant. If it varies, then it cannot be used as known-plaintext.
But then you still have the problem of identifying the contents. If there were no headers, one could not tell if the message was compressed using ZIP, LHA, StuffIt, tar*, compress, gzip, Alice's Magical Supercompressor, or even if it was left alone. One could also not tell if the decryption happened successfully. ( * Yes, I know tar is not compression. )