Perhaps the incentive is for variable behavior. When there is low GPU demand, burn more, but reduce when there is contention.