Cool! Let us know how it goes OC.
Since I don't have a quad running XP, I'm interested in seeing the results. It could help many here. Here are the things I'd like to see tried out:

1) Pairing each SMP instance on one die (share an L2 cache):
Code:
C:\Folding@Home\Windows SMP1\FahCore_*.exe := PAIR0 [assign=1,resource=CPUUSE,policy=PSEUDOBALANCED]
C:\Folding@Home\Windows SMP2\FahCore_*.exe := PAIR1 [assign=1,resource=CPUUSE,policy=PSEUDOBALANCED]
3) Now sprinkle in the GPU clients (one client per pair)
Code:
C:\Folding@Home\Windows SMP1\FahCore_*.exe := PAIR0 [assign=1,resource=CPUUSE,policy=PSEUDOBALANCED]
C:\Folding@Home\Windows SMP2\FahCore_*.exe := PAIR1 [assign=1,resource=CPUUSE,policy=PSEUDOBALANCED]
*\Folding@home-gpu1\FahCore_*.exe := PAIR0 [force=yes,priority=belownormal]
*\Folding@home-gpu2\FahCore_*.exe := PAIR1 [force=yes,priority=belownormal]
2) Distributing the busiest PIDs from all clients across all cores:
Code:
*\FahCore_*.exe := ALL [assign=1,resource=CPUUSE,policy=PSEUDOBALANCED]
Hmmm this is tricky without testing it. Not sure if I have all of this correct