This is not about being genius. Terms of purchase of fpga and efforts to implement the project give different results to different people. That is all.

Regarding x16, x17 and xevan. I would suggest to use arrays of folded cores. Those folded cores are fast and small. It could be possible to make arrays of cores for all SHA3 candidates; balance those arrays to mantain the same hashrate; make a mux  in the front of cores.
Probably the monstrous setup with gigabit transceivers is not needed.