Thanks for you comment. Yes, with PX it is possible to have many more than 2 concurrently active sorts on behalf of a user session. However, the PX slaves each execute slave SQL in separate Oracle sessions, and each slave session can have at most two concurrently active sorts. The point that I was trying to make was that the "worst case" is not sort_area_size * processes as is commonly thought, but (sort_area_size + sort_area_retained_size) * sessions. The fact that a parallel query may invoke multiple slave sessions does not change this "worst case" formula, it just increases the number of active sessions.

I think there may be a mistake in the sort memory article. In actuality, more than 2 sorts can be going on simultaneously for one session if you include parallel queries. If I have a DOP (degree of parallelism on a query) of 10, and I'm sorting the results, I can actually have up to 10 sorts running simultaneously for that one session. In fact I'll get 10 PQ slaves for retrieving rows, and 10 more for sorting purposes.