Apparently it tweaks some sched params for better user experience...
We pass max_loop=255 via kernel argument, so for this to work the loop device must be compiled in.