> What if statement caching is used? Does it release by "closing the
> statement" in that case (I am using a connection pool as well so
> they never officially "close" underneath).

The main problem is that when you close the result set, the corresponding
"fetcher" is not closed. In autocommit case corresponding fetcher contains
complete result set. When you execute the statement again, the fetcher is
replaced, so the previous is garbage collected.

In the non-autocommit case the fetcher is also referenced, but that is
relatively small object.

> Just wondering how severe this problem is (and if I should upgrade
> as soon as the next release comes out).

Depends on the usage scenario and result set sizes.

Roman