The optimal policy for load-leveling in local area networks have been depends critically on the nature of the workload, but little attention has been given to educational LANs whose workloads are atypical because their main purpose is to control the distribution of software. The authors review the characteristics of this class of network, where response time is the dominant criterion for judging policies for allocating tasks among servers. A study of the response times of various policies through a combination of closed-form queueing theory models and discrete-event simulations revealed that cache memory in the server was the critical bottleneck and led the authors to select one of these policies in the expectation that it would improve response time substantially. Experience has confirmed the authors' predictions.>