getting our bearings on "out of memory. failed on request of size..."

Hi All,

   Our PG server is serving up 50 databases or so (with identical
schemas), and the largest one of those has started giving us "out of
memory. failed on request of size n" errors in many places. This is a
9 GB database with 100+ tables, the largest of which have 4 million or
so rows. We see the Out of Memory errors when trying to run a "hard"
SQL statement (SELECTing from a complex view, updating all rows in a
large table, etc...)

   Can anybody point us in a good direction of how to track down just
why this is happening? Is there something we should be looking for in
the log (and log settings to turn on)? What is a good way to log
memory usage of statements?

   Any help much appreciated in figuring out how to approach problems
like these.

Thanks,
-Tom