These are very interesting questions. I like the idea that the physical work of encryption and decryption might be accounted for by the increased entropy of the message.
Is this affected by whether or not the key is known?
If the key is unknown, then the meaning or affect of the message on the recipient is that of noise (or more accurately, "this message exists but you can't read it"), but information theory has nothing to say about this since it is concerned only with the communication itself. Perhaps the AI theory of Language As Action could be connected to information theory (yeah, recast meaning as communication to a homunculous...) Somewhere I have a paper on Maxwell's Daemon and Data Compression, which seems related to this. I can dig it up if you want a ref. (I haven't read it yet; it's ftp available.) Paul E. Baclace peb@procase.com