Thank you so much Suvashish and Jeff for your replies and assistance.

We managed to isolate the problem. It was an intraweb application that we had running which somehow kept grabbing memory. It has since been taken off and things are back to normal.

I shall study the metalink document closely nevertheless as there's definitely much more we could do to configure the sga and share pool properly. Strangley, Oracle support failed to point us to this document.

Thank you again, one and all!