MPI GPU CLusters

I am building a rather large cluster of Jetson TK1 development boards and I have been investigating a means to implement CUDA based MPI and I found some interesting information: https://developer.nvidia.com/mpi-solutions-gpus

I started wondering how difficult it may be to use this with oclHashcat.

Does anyone else find the idea of distributing a workload across several GPU's in a cluster fascinating?