Have you guys looked at Cuckoo cycle?
FWIW we evaluated Cuckoo as well for Zcash, and it was a strong second-place contender. There wasn't really anything wrong with it it just didn't seem to have quite as much of a rigorous scientific analysis as Equihash. However, that is a very subjective thing for me to say. You could argue (and Cuckoo's author, John Tromp, does argue persuasively) that Cuckoo's history of analysis and refinement is better than Equihash's.
What about cycling through 10 unique PoWs every 10 blocks?
I'm not the best at discrete analysis and understand this multiplies attack surface 10-fold, but could we splinter miners into small, specialized, and de-fanged factions using 10 different well-chosen hash algorithms, then scatter them among CPUs/GPUs/FPGAs/ASICs?
Block 1 JH
Block 2 Skein
Block 3 Groestl
Block 4 Cuckoo
Block 5 Keccak
Block 6 Equihash
Block 7 BLAKE2
Block 8 SCrypt
Block 9 CryptoNight
Block 10 Ethash
DeSantis has started some work (he wants to do some testing before posting his source code for peer review though).
He's creating a Keccak fork and a Cuckoo fork, and has created a beautiful automated testing utility that I hope he gives me permission to link to you guys.
The testing utility (I've viewed the source, it's not vaporware) allows you to spin up multiple Docker containers, each containing a different Bitcoin node; some of the nodes can be Bitcoin 0.14.0, some of them can be Bitcoin Unlimited, and some of them can be Keccak, Cuckoo, etc.
With these containerized Bitcoin nodes, you can then simulate various forking scenarios, and actually observe in real-time how it plays out. With my limited bitcoin programming knowledge, I am waiting for him to document the config file that controls the node counts & types, and to create some python installation script (which are easier to debug for me at least).
tl;dr -
DeSantis is testing Keccak & Cuckoo using a Bitcoin Network Simulator.