Is the limited entropy of the shared secret due to the fact that it is simply a hash of a password? If so, then perhaps the current simplistic shared-secret key protocol is not such a good fit for these common shared-secret password applications.
The fact that most users wont have a shared secret with 256 bits of entropy? I suspect that most shared secrets are probably in the 64-80 bits of entropy at the highest, and probably much lower than that.