In theory it would be nice - restart performance would increase, various glitchiness related to pods moving around would get reduced, could have distributed disconnected nodes that only occasionally connect to control plane, could have some basic form of persisted storage...