A cost-effective approach for enhancing availability of clustered system is to cache popular data across the memories of servers so as to alleviate disk I/O bandwidth bottleneck. The cache placement problem is complicated for the fact that video popularity changes over time. Moreover, as a new multimedia streaming service, time-shifted TV introduces new challenges to the problem. In this article, we investigate the crucial problem of the clustered time-shifted TV system. We formulate it as a multi-objective optimization problem based on the time-based popularity model. We propose a classification replication algorithm and a novel cache placement algorithm, named as cost-aware least load proportion first (CALLF). The algorithms focus on the time-varying nature of time-shifted TV, and explore previous stored information to reduce the cost of redeploying. We also present a dual-threshold reconfigure scheme to further balance the load of servers dynamically. Simulation reveals that the proposed algorithms are efficient on improving the performance of time-shifted TV system in terms of total outgoing bandwidth requirement.