This seems like a pretty common problem for people using the postgres backend. If we're going to release 1.4 sometime soon I'd rather see this included as-is than wait for a better version that may not materialize :)

That said, I've attached (a completely untested; I'll try that tonight :) version of Erik's patch that pushes this more into the Postgres backend. The PG backend provides its own getSessionManager() method that returns a custom Session object that handles the try/except during a set. Erik, do you think this approach is a better candidate for inclusion in HEAD?

If so (and provided my testing goes well), I'll commit it (or something like it) in the near future.
File Added: back_postgres.diff