Consider a set of discrete tasks whose stochastic task times have specified means and variances, and these tasks have to be assigned to one of the two serial stations of a flow-type production system (e.g., an unpaced line) with an interstation buffer. The present purpose is to investigate: (i) what ideal combination of means and variances of station processing times will give the highest system utilization (U); and (ii) how serious it is to deviate from this ideal arrangement. A literature review shows that earlier related results are not applicable due to their modeling restrictions, and simplistic extension of earlier results leads to contradictory conclusions anyway. Numerous simulations were made for the system with different buffers sizes and different combinations of station-processing-times’ means and variances. Conclusions from analyzing the extensive simulation results are: (i) U is maximized when the means and variances are both balanced, i.e., the two-station-processing times have the same mean and variance; (ii) variance imbalance has a surprisingly small effect on U, but this effect is greater under greater mean imbalance; (iii) for a given mean imbalance, a much larger counter variance-imbalance is needed to achieve a best possible U, but this counter variance-imbalance can never fully compensate for the ill effect of mean-imbalance on U; (iv) the ill effect of mean imbalance on U decreases substantially if the discrete tasks have more variable task times.