Most Video-on-Demand scheduling models are focused on the optimization of disk and network utilization, paying less attention to memory management. In this paper we present, analyze and evaluate a new buffering policy, namely Single Pair of Buffers, which greatly reduces Video-on-Demand server memory requirements when used in combination with any cycle-based scheduling algorithm. This technique is based on the synchronization of a single pair of buffers per storage device, as opposed to existing solutions based on per-stream buffering allocation. The influence of factors such as Variable Bit Rate streams, zoned disks, network speed and cycle length has been considered in detail. Extensive simulation results show that Single Pair of Buffers leads to more than one order of magnitude memory reduction when compared to the popular double-buffering approach, at a cost of a minor (less than 2% in the worst case) performance degradation.