Revisit internal pipeline system

[mirceaulinic]

We need to investigate more around zeromq/pyzmq#1009 in terms of latency and throughput over the IPC vs the multiprocessing Pipe.

I personally still believe that zmq should perform better. I also suspect that Pipe may miss messages when the buffer is full (send does not fail, hence we can't know whether the message has been queued or not). On the other hand, zmq does not miss any messages for sure, but we need to dig more into the delay issue.