Did they evaluate good old bcrypt? I haven't looked at it in a while, but it used to be very GPU unfriendly (though still vulnerable to FPGA/ASIC).

This seems like the real mechanism here is not actually proof-of-work so much as security-through-obscurity.