Hybrid Myrinet programming (SMP in clusters)

Hi everybody:

I am going to work on a cluster with a few (2-3) SMP, namely dual Xeon. I have 
seen somebody is attempting to modify MPICH-GM so that it works in both 
shared and distributed memory environment. My current plan is to use GM or 
MPICH-GM with OpenMP. In any case, the program is going to be compiled using 
Intel Compiler, because I need good, processor specific performance optimization. 

By the way, every computer in the cluster is using Windows XP (required, not my 
choice), but I am open to any suggestions.

Thank you very much.

Wilson