This should still be fairly parallelizeable, shouldn't it? You go ahead and take a number of crackers. Each cracker starts with a random hash and randomizes it

This should still be fairly parallelizeable, shouldn't it? You go ahead and take a number of crackers. Each cracker starts with a random hash and randomizes it for a fixed number of times, or until the bottom 8 bits of the hash are 0. Once it is done hashing, it goes to a central database and stores it's hash sequence. After that, it picks another hash which is not contained in a hash sequence and repeats the process. Finally, you'd probably have one node just doing the boring hash iteration while checking if it can skip according to the generated hash sequences from the database. I haven't done any math how much of an improvement this would be, but I would suspect a pretty decent improvement given enough nodes.