In current Pyphant, calculation is done multithreaded, which does not scale on multi-cpu because of the GIL.
I'be built pms/pypubsub, (pyubsub.sourceforge.net) which essentially makes it possible to exchange python objects between processes, in a publish subscribe manner.
I've attempted with this framework to create a distributed data flow calculation example, and it turned out to be possible, easily!
I've not looked at the pyphant core, but I think it is possible to use the pms framework for distributed calculation for pyphant.
If you like to discuss, or interested, leave a note in the forums on pypubsub.sourceforge.net, and I will send you a piece of example code showing the principle.
Thankx
Carl.