Horizontal scaling is key. If we had more faith in auto-sharding in MongoDB 1.6, that may have worked (however horizontal scaling wasn't our only issue with MongoDB).
With PostgreSQL we manually partition the data (luckily we have a really nice key to partition on).
PostgreSQL 9's streaming replication has been adequate so far. We've only been using it in production for a few months and recently moved off of AWS* which drastically reduced our error rates overall.
* http://urbanairship.com/blog/2011/06/01/hybrid-cloud-were-up...