We extend the validity of some results on the optimal control of two-server queueing models with service times of unequal distribution, operating in continuous or discrete time. The distribution of arrivals can be arbitrarily subject to some conditions. Both discounted and long run average costs are considered. Dynamic programming and probabilistic arguments are the key tools used to establish the assertion that the optimal policy is of threshold type, i.e. the slower server should be utilized only when the queue length exceeds a certain threshold value.