Otherwise we will emit buffering messages not just from the last multiqueue but also from previous multiqueues... confusing the application with different percentages during pre-rolling.