The only previous work on the subject is [33,32] that partitions the cache into a ``hot'' zone that is managed via LRW and a ``cold'' zone that is managed via LST. The authors point out several drawbacks of their approach: (i) ``The work in this dissertation only deals with the interaction between the cache and one disk.'' [32, p. 126] and (ii) ``One of the most immediate aspects of this work requiring more research is the method to determine the size of the hot zone for the stack model-based replacement algorithm. We determined the best size for the hot zone empirically in our experiments.'' [32, p. 125]. To continue this program further would entail developing an adaptive algorithm for tuning the size of the hot zone. However, note that the hot zone optimizes for temporal locality (say ``apples'') and the cold zone for spatial locality (say ``oranges''). It is not currently known how to compare and trade apples versus oranges to determine the best adaptive partition. In this paper, we will present a natural combination of LRW and CSCAN that obviates this need, and yet delivers convincing performance.