Couldn't you do something like this?
The sequencing computer sends midi to each of the GS slave computers.
Each of the slave computers then sends their audio to the gigapulse computer, which feeds the finished result to the sequencing computer.
If this is what you are thinking of, it is certainly possible and doable with some caveats.
The gigapulse computer if it has to provide convolution for half a dozen or more giga slaves will need to be extremely fast - something like a quad opteron or xeon.
It will also need a GSIF2 soundcard and enough inputs (analogue or digital) for all the slave computers.
Presumably, you won't be bothered about the extra latency.
But at least the latency will be much lower than using a switch or router to link all the computers.
I think the more usual way to dedicate a machine to GP would be to treat it like any other dedicated reverb box. Feed the audio to be processed into it and return processed audio. Perhaps with adat ports if you want the possibility of 8 channels of processing.
With GigaPulse, I think the bigest issue will be how many instances your dedicated machine can process. I suspect the Teleport and gigabit Ethernet processing might cut into the limit. Sticking to audio processing with audio cards would certainly get you OS independence... your DAW machine could be anything, even a digital mixer.