Using trace driven simulation, we compare the performance of GreedyDual-Size with LRU, Size, Hybrid, and LRV. Size, Hybrid, and LRV are all ``champion'' algorithms from previously published studies [WASAF96, LRV97, WA97]. In addition, for LRV, we first go through the whole trace to obtain the necessary parameters, thus giving it the advantage of perfect statistical information. In contrast, GreedyDual-Size takes into account cost, size and locality in a more natural manner and does not require tuning to a particular set of traces.