That's pretty cool...
One of the hole cards held by Postgres is that the traditional users in past times were using it to hold things like astronomical data. Single or a few users using a massive database in a lab.
So while PG has historically had problems with reliability, with being very slow on simple selects (always fsynch'd the log, ugh, a killer for websites!), etc it has always been capable of handling large datasets. Big tables get split into multiple files automatically, etc etc.
So I'm not terribly surprised at the success story you've posted. At the moment, PG's main problem when one's running really huge databases is that there's no in-database support for placing tables and indices in separate filesystems, for instance onto different drives. You need to move files by hand and soft-link them. This is admin-unfriendly, and will change, hopefully in PG 7.2 ...
