PG doesn't implement two-phased commits, though there's been talk about doing this in order to support true replication throught the write-ahead log.
In the interim, there's a set of master-slave replication scripts that supposedly work fairly well. They were written by the guy who added write-ahead logging and row-level locking to PG, in other words by someone who knows what they're doing. However I have no idea if this approach is robust in the face of the slave going down (it may be, I just haven't looked at it). The basic synchronizing work is open sourced, but the admin tools which lay atop it are sold as a product by PostgreSQL, Inc. They've promised to open source these as well in a year or so.
It's worth investigating, at least...
