The quantum scenario is not particularly applicable here. Generic preimage attacks on sponges are essentially collision-finding on the capacity. Grover does not

The quantum scenario is not particularly applicable here. Generic preimage attacks on sponges are essentially collision-finding on the capacity. Grover does not beat classical algorithms at this; Brassard-Høyer-Tapp doesn't either, if you take into account the full cost of the algorithm. In any case, having a c = 512 ceiling for the capacity would put every attack at over 2^256 cost, which presumably is enough, and would keep the function reasonably fast. Note that c = 2n would have rivaled Skein and possibly BLAKE in terms of speed. As for analysis, what matters is the analysis performed on the permutation. The permutation was not touched, and touching that would indeed raise alarms.