Well, you probably shouldn't cluster on a UUID, but it's not really a great idea to blindly cluster on a sequential id, either. You should cluster based on how you query the data to minimize I/O.