This helps performance, as the thread creation overhead in Linux is ~10us. This also helps profiling, since the threads which handle a certain task have a consistent thread id. This also helps CPU logging have a smaller number of stacks to manage.