Why stop at 2 processors - or 3?
I would be interested in the possibility of Csound running on a large array ,
where issues like synchronisation and the allocation of differing processing
loads (running Csound sub-tasks) were dealt with by the system itself.

I might be very wrong, but aren't these the sort of things that arrays do well?
 
Perhaps the Parallel Real-Time Systems and Speech Processing & Acoustics
Research Group in the Dept of Electronics at University of York can shed some
light on the idea of Csound and arrays.  
One of their number - John Tuffen - is/was working on something last year
called "Dynamic load balancing on systolic array processors for use in music synthesis".

See www.amp.york.ac.uk/external/aseg/restaff.htm for further details.

Can anyone tell me how an array of processors differs from a neural net?