On optimal delivery lags