My opinion/sense is that Postgres is better in every way, except replication. Which ironically makes it far worse than mysql since you can't easily use it for huge datasets.
Replication does nothing to enhance searches over large datasets, and in the case of master-slave replication, you don't even have a second server to query.
Not to mention, that MySQL replication is known to be horribly fragile, and silently corrupts data -- which is why you need separate tools (like Maatkit) to know when it's time to rebuild the slave.
There are several replication solutions for PostgreSQL but they all require a level of sophistication and configuration from the administrator. One of the biggest issues I have seen in the field with setting up replication for postgres is that people design databases without primary keys and with duplicate rows. A DBA who understands the relational model wouldn't do that in the first place; and none of the common solutions for replicating a postgres database will let you copy a database with broken referential integrity.
No kidding. Why are people so surprised that using an RDBMS well requires an understanding of databases? Database administration requires a very different skill set than programming, and DBAs have different priorities (e.g., Do NOT corrupt the data. Ever.); they're not just there to inconvenience programmers.
It seems like almost every time I've read a review of a database book, there's been somebody complaining that it's "full of a bunch of boring theory" (or else, "at least this book doesn't have much boring theory"). At the same time, some programmers nonchalantly do things that could land them on a DBA's equivalent of The Daily WTF (e.g. http://ask.metafilter.com/117908/Theres-got-to-be-a-faster-w...). It's a weird disconnect.
Well, there's a couple of issues out there; one being that most of the DBA's I've met were coders before they specialized, so on average most DBA's are more experienced. Another is that a shocking number of people in the industry don't understand and don't want to understand the basic concepts underlying the field. There is an unfortunately large class of developers that seem to want a pile of Lego Bricks that they can assemble into a semblance of functionality and who don't want to "bother" with understanding the fundamentals.
Slony-I is a terrible piece of software; it's incredibly difficult to administer, requires full table locking across the entire data set (EVERY SINGLE TABLE!) to do DDL changes, and occasionally breaks in mysterious ways which can lead to data corruption.
I haven't used PGCluster, but it's synchronous multi-master and thus inherently fails to scale out as well as async systems like Slony-I and Londiste. Probably great in circumstances with only 2-3 db boxes though.
Londiste, on the other hand, is awesome, and with the 3.0 release now has hot backups. Simple to administer, rock solid, and has really important features like live DDLs. Demand Londiste and the SkyTools package! Don't settle for less!
