Meeting the service demands from QoS-based network applications is a very challenging task performed in many high-end routers and switches. This task involves management of resources like bandwidth and memory in network devices. The memory in the form of a very fast cache that instruments wire-speed classification, discrimination, and forwarding of network packets needs to be managed very effectively. We examine the management of a specific IP flow-cache architecture through simulations based on traffic traces collected from a campus intranet. A probabilistic cache install policy is examined over a range of cache sizes and install probabilities. This policy successfully identifies the flows that warrant caching and slightly improves deployable policies based on site-specific traffic information can increase the switching performance even higher.