I have been looking into this topic myself for some time Stripes, I have identified a few options that make it possible to utilise the existing networks infrastructure to carry new method packets inside older message formats, of course I,m not going to divulge those here but they do exist.
There are currently two lines of thought on the encrypted file transfer option, one is to add this to the patch so its done outside of mx and mx never sees the encryption layer or the other is to add such features to a new client , as you know that is not something we have in hand atm and no doubt at such a time we will have a whole new set of ideas to chew over
The basic operation of winmx is to take 2k audio or 4k vid chunks at a time in sequencial order
this is great for previewing material but useless for beating a throttle, creating a temp array to reorder the segments would be a small help for privacy but still leaves one major aspect that I havent manged to theorise around completely yet, the size of a file is not often something detectable by the throttlers but the type, port and duration of the transfer are what trigger the throttling, so just creating a solid wall of encrypted traffic may mean we will still see no positive effects.
We escape much of the heavy bit torrent throttling due to the limited number of concurrent nodes in use at any time however we need to ensure that we cover as many of these bases as possible in a mutatable (or psuedo mutatable) fashion.
If you have a theory you want to put forward or need more info on the existing method of transfer please feel free to either discuss this here or join the core developement section and post it there where I think it will reach the intended audience.
Let me know if you wish to join that section as its invite only and drama free so far.