The drawback of the approach of assigning a given processor at each time step to work on the same layer is that all the processors would have to read new data, and the overhead caused by this input operation could void all the benefits of using parallelization in the first place.