What hardware configuration makes it practical to move UBB.threads past the "800-1000 users online" level to something say 5x that and perhaps even beyond. Is there an understood and documented way to get this performance?

I am open minded about configuration: big SMP server, many smaller machines, solid state disk, etc. (Presently have 30k members, 40k posts, and 800 users online in 10min window at peak on 2x AMD64 w/2GB RAM.)