The thought occurred to me while reading Matt Welsh’s criticism on how Project Graffiti conducted their experiment. I think it is possible to share files over twitter and I will sketch the concept bellow:
- Create N accounts on twitter.
- uuencode the file in question so that we have to deal only with printable characters. Optionally encrypt it first.
- Pick randomly the account that will tweet first.
- tweet the first line of the file.
- Use a hash function (like SuperFastHash) and hash the line. That way you will get the next account to tweet the file contents.
- The next account replies to the previously tweeted line. The reply contains the next line.
- Repeat hashing and replies until the whole file is tweeted.
When the process is finished you end up with a string of replies that if put together contain the (uuencoded) contents of the original file.
Yes, this is neither practical nor distributed file sharing, but in a way it is hiding obscuring information in plain sight.
Managing organized complexity 
or just use tpb
This idea was not meant to “compete” with TPB. Rather, it was a concept similar to using DNS caches (and TXT resource records) to share CD-ROM images.
This kind of file sharing is “pass by value” instead of “pass by reference”.