One important feature for PostgreSQL which they should have mentioned is transactional DDL. It makes applying migrations much less risky since you know if a statement in the migration fails all of it will be rolled back.
After having used PostgreSQL in production for some years I do not know how I could go back to a database without transactional DDL.
Awesome. Given that their services look Ruby-centric, maybe they could take over PL/Ruby and point out that this would allow people to write their stored procs in Ruby too.
I can't give you a comparison, but for what it's worth, it is used on the Postgres documentation and mailing list archives. I mention this because you can play with it and get an intuitive sense of its performance and subjective usefulness.
It's also handy because you can combine it with relational data. For example, if you search from a manual page of a particular version of Postgres, it'll only return results from that version.
The main performance limitation is probably update/insert rate, and it doesn't has as many features as a dedicated IR-type system, but it's frequently good-enough and already present.
Postgresql cannot be fast as Lucene because of its not a "search engine". Postgresql full-text is fast enough for most cases but it doesn't have support for facet queries which is important for a search engine.
Actually, it does, which is why tools like PostGIS are built for it. In the most recent versions of PostgreSQL (I believe 9.1) they even added k-nearest-neighbor multi-dimensional index support to GiST.
Postgresql is nowhere, NOWHERE near as powerful or as fast, but it'll work for the cases where you're doing simple search-interface bolt-ons to an existing product already using PGSQL.
Postgresql fulltext and ElasticSearch are ultimately solving two very different sorts of problems.
The only case I can think of where you would prefer to use PGSQL is if you some weird blending of normal relational data with the documents, but I think realistically you can sidestep that. We've been able to do pretty complicated queries against our ElasticSearch instances.
As you say they solve two very different sorts of problems.
If you want to do full text searching of documents, probably PostgreSQL is not the right answer. If you want to pull all customers in the UK with notes searched full-text-wise for "did not pay" then PostgreSQL is much better just because it all integrates into the same filter.
After having used PostgreSQL in production for some years I do not know how I could go back to a database without transactional DDL.