This paper considers optimal design models for queuing systems in which the decision variables are the number of servers and the mean rate at which each serves, the total cost per unit time of operating the system is the sum of a service cost per unit time and a waiting cost per unit time, and the optimality criterion is long-run expected average cost per unit time. It is shown that a single-server system is optimal for a wide class of arrival processes and service-time distributions, and for a wide variety of waiting-cost functions.