Many manufacturing systems can be modeled by multiclass queueing networks. A semiconductor wafer fabrication facility is an example of such a system. Some servers, like furnaces in a wafer fab, may be able to process a batch of jobs simultaneously. Some servers may need a significant setup when they switch between job classes. Myopic scheduling policies like waiting-for-a-full-batch or avoiding-a-setup are tempting for potential high utilization of the network resources. However, they may be inefficient in the sense that bottleneck resources are underutilized no matter how high the work-in-process is. In this talk, I will discuss a number of scheduling policies that have been proved to be efficient. The policies are generated by studying the corresponding fluid models of queueing networks.