Upped memory guidance

[Nate Klingenstein]

> It's pretty unlikely to go unnoticed because of the amount of memory needed, it's more likely to just fail than keep going. Mine certainly did (after increasing the amount).

That was my reaction, too.  I've killed off plenty of IdP's by running them out of memory, but typically through concurrent sessions, a load test, and "cozy" VM RAM.  That was java failing to grab more spare memory from the OS, which might manifest itself less elegantly than specific methods that hit the Xmx cap.

The report I heard was 2.4.4 under moderate load but no Terracotta or similar. Dunno about bitness.