There have been a handful of benchmarks that show that Postgres does remarkably well in comparison with Oracle in raw transaction performance.
I've not done any serious benchmarking but I work in the OpenACS 4.x framework with developer support enabled, and when comparing pages loading under the Postgres or Oracle versions find that the database queries are reported as taking roughly the same time, with Postgres queries generally being very slightly (10% or so) faster.
However, this is on a mostly empty installation.
Something to remember is to vacuum analyze your installation just as soon as you install it (hmmm...maybe I should have the bootstrap/installer do that automatically) - it makes a big difference. Oracle seems to do a much better job without statistical information on table contents than Postgres.
Postgres also degrades in situations where you have a very high load of inserts and updates. Nightly vacuums are sufficient for most sites, but can take quite a long time, which is a bit of a drawback for 24/7 web sites visited by people all over the world. PG 7.2 will include the option of vacuuming in the background or at low priority, continuously (or more or less so), alleviating this problem.
The short story is that many of us find Postgres perfectly adequate for our web/db sites.
openacs.org - the site you're looking at right now - runs on Postgres and is solid and runs months at a time with no human intervention.
