OK, to overcome the spam, I'll post my opnion on these three:
Oracle: Large, mediocre performance for single users, but scales to huge heights on big hardware for many users. Every feature plus a few you've never even thought of. Designed with the enterprise in mind, and it shows. When downtime costs you $1,000 a minute, it's easy to understand why people pay $20,000 a year for a CPU for the base install. Lost of commercial applications are written to use it. new versions are known to have issues, which is why most big companies are still installing version 9. Achilles Heals: Price, quotas are a serious PITA.
PostgreSQL: Most active / capable development community around. Huge strides made in the last few years, with things like point in time recovery. User extensible means that new features can be added by a team of one with no intervention required by the core developers to accomidate such changes. For example, most PLs were implemented this way, as were slony replication, PostGIS, and tsearch2. Releases new versions about once a year, so the version you tried 1.5 years ago is likely considered out of date by two versions now. New versions tend to have minor but bothersome bugs that are worked out in about 4 to 6 months. Very active mailing lists, with the primary developers involved daily. Achilles Heals: Vacuuming (it's getting better), indexes other than btree tend to be red-headed step children in some ways (not WALed, slower, buggy), finding and installing expansion items like jdbc or plphp can be a bit of a pain.
MySQL: Insanely popular and for good reason. it's easy to install, easy to use, and easy to administer. While it has serious weaknesses when used in the role of an accounting database, as a content management database, with few writes and lots of simple selects, it really shines. Very fast turnaround on bugs (I reported one in 5.0.3 that was fixed in <24 hours). It's less modular design (compared to postgresql) means that large features require more development coordination to ensure nothing breaks unexpectantly, seems to fit well with its more commercial development team. Achilles Heals: Table types other than MyISAM don't get as much testing therefore may have corner case problems that haven't been found, can't set a default table type, silent data corruption (i.e. silent overflows, partial transaction rollbacks due to table type mismatch, etc...), documentation not broken out by version.
That last one is my personal pet peave, and drives me nuts. I hate reading three pages of docs, two of which don't apply to the version of whatever product I'm using.
