We have built our cluster with the MPI implementation of John the Ripper: Running bcrypt hash on 32 Zynq chips (24 Pynqs and 8 Zedboards). Our cCc (cool_Cracker_cluster) computes 140 k hashes/s at nearly 155 Watts: Quite impressive if you compare it to Nvidia Tesla K100 (55 k hashes/s, so 2.88 times faster), short video available here.
If your are interested, more to follow in the upcoming technical report.