The communication-storage tradeoff, as a key performance metric of the fundamental limits of caching, has attracted considerable recent attention. In this paper, the issue of how much storage cost should be paid for a target effective throughput is investigated in a unified framework. This approach, from a queueing theoretic perspective, adopts Little's law to analyze the average buffer consumption, thereby giving a rate-cost function that relies only on the probability of content request delays. A time sharing policy along with its optimality criterion is further proposed to achieve the optimal storage efficiency. For pushing flows with heterogenous request delay information, a joint cost-rate allocation method is presented to maximize the overall storage efficiency in either a centralized or decentralized manner. Both analytical and numerical results reveal that the storage efficiency of caching is dominated by the demand probability and the maximum request delay.