Anyone able to record with BFD All and HW buffers of 128?

[j20056]

I am contemplating purchasing a super fast multiple core machine, as I simply cannot run BFD All succesfully in a recording environment. Specifically, I have been having a few messages with Rail at Platinum Samples, from whom I bought the great sounding Andy Johns drum samples for BFD.
The problem I have on my current setup (PC 3GHz Pentium Duo-Core - 3GB RAM) is that BFD uses 45% of my CPU power, and I must run at 512 buffer size in order not to get the usual "H/W increase buffer size" error message. Unfortunately, 512 buffer size leads to unacceptable latency.
What Rail said is that he runs at 512 buffer size. He also said that many record with say a Roland TD drum module to get the MIDI, but they don't use BFD during recording, only after. During recording, they use the Roland or any other zero-latency drum external module, then do edits and bounces at 512 buffers, wh4en latency is no longer a problem. I really would prefer a low latency BFD based recording process, so my drummer can hear exactly what gets recorded, and ideally at 128 or lower buffer size.
has anyone succesfully integrated BFD in an actual recording, and not just editing work mode at 128 buffers or lower? Is this possible with an Quad or Eight core machine? Is there a machine by a leading manufacturer (sorry, I can't build an Allenstein Quadzilla meachine myself) that could achive top performance of that type?