DJB posted to the NIST SHA3 mailing list a couple days ago suggesting fixing R at 1024, giving the hash a capacity of 576 and a security against preimage attack

DJB posted to the NIST SHA3 mailing list a couple days ago suggesting fixing R at 1024, giving the hash a capacity of 576 and a security against preimage attacks of >= 256 bits in all configurations. This sounds pretty prudent. It avoids any security gotchas, and will also make life easier for implementers because the hash will consume a nice round 1024 bits of data at a time for all hash sizes. This config would have slightly greater anticipated security at 256 bits than the submission, somewhat worse at 512 bit output— but in both cases be better than NIST's proposed parameters... and the performance impact would be relatively modest.