Try cutting pga_aggregate_target in half. Have you checked the 9206
patch set to see if the pga issues it fixes apply to your environment?
I've been convinced (without proof) it doesn't work as advertised, just
blindly cutting it seems to fix problems (for my 9206 system, 400M
works good). But see
http://www.jlcomp.demon.co.uk/faq/pat_presentation.html and
http://www.jlcomp.demon.co.uk/untested.html for real information.

Do you have multiple instances? Shared servers? How much physical
memory do you have? What are your init.ora parameters that affect the
SGA, anyways? sort area size?

Some of the recommended unix kernel parameter settings have changed
over time, have you checked them lately? If you are not running out of
memory, then you may be running out of a memory resource that is
dependent on the kernel parameters. I assume you've seen Note:233869.1
and Note:199746.1, but go over them again anyways. What's your
unlockable_mem, too? (This can get screwy if you let the system manage
it, especially if you've got a large SGA, or bogus shared memory
segments. The system can't lock memory for a process, even though
there appears to be plenty of memory available, so gives back a message
to Oracle, which then 4031's).

jg
--
@home.com is bogus.
RAC really doesn't like being bored:
https://metalink.oracle.com/metalink...,1,1,helvetica