In our evaluation we quantitatively describe the magnitude of the overheads that data replication can introduce as well as the benefits of efficient recomputation. RCMP’s benefits hold across two different clusters, with job inputs of either 40GB or 1.2TB. RCMP is implemented on top of Hadoop. In our experiments, during failure-free periods replication is 30% to 100% worse compared to RCMP. More importantly, by being efficient under recomputation, even under single and double data loss events, RCMP yields better or comparable total multi-job running time.