I'd say your tx settings are fucked up, or you misconfigured your CMP engine.
I am actually experiencing the same problem the previous gentleman experienced. The query time with 4 concurrent users is around 4 minutes. Obviously, I must have f'd up something. But could somebody give me some pointers on how to configure the tx correctly to cure this problem, or simply where to look at to improve the performance?
It could be that serialization is killing you. You can test this by retrieving the resultset, serialize it to a file (or what ever) and read it back again checking the times spend along the way.
You can speed things up orders of magnitude using the --Externalize-- interface.