Roberto and I have discussed this in private e-mail, and I've asked that he try to bundle up a reproducible example to shuffle off to the PG group.
In reality, one normally would do a dropdb before a full restore just to clean out the various tables. I'm also curious as to whether or not Roberto got any errors during his reload (Roberto?).
Anyway, it sounds like one or more system tables got screwed up by the reload, and even if reloading fails it shouldn't cause backend crashes.
