[Neil wrote regarding MySQL performance:]
In the 10% write case, however, Postgres kept going beyond the client-concurrency load where MySQL falls over.
Personally, I suspect the hit-tracking feature is responsible for much of the slowdown. Currently *every* normal page load causes a write to update the number of accesses ("This page has been accessed 281 times.")
I recommend making the hit-tracking feature optional in the code, then turning it off for a few days.
In the long run, this kind of per-page tracking could be done more efficiently using the webserver's logs, perhaps updating the access count once per day using a separate script.
--Cliff