>I'm testing 50 elements as I write. So far 12 minutes on a PIII 600 with 384 MB ram.

At a glance, this is just about the same "combinatorial" problem that we had last week, isn't it? In that case, with 50 elements, you will have this many solutions:

2^50 = 1,125,899,906,842,624 (= 1.1 quadrillion)

Not nearly as bad as the 2^70-1 solutions of the other problem, but still, it could take some serious runtime...