And doesn't processing necessitate buffers? How does this approach reduce the number of buffers involved? If anything, it sounds like less sharing would mean more buffers.