A data recovery to maintain redundancy from multiple node failures often occurs in dynamic peer-to-peer storage systems. Most of the existing recovery schemes are designed to repair these node failures one by one. In this paper we design a mutually cooperative recovery (MCR) mechanism for multiple node failures, that is, new nodes for recovery are repaired mutually cooperatively. Via a cut-based analysis in information flow graph, we find out a lower bound of maintenance traffic based on MCR and prove the existence of a random linear scheme satisfying the minimal bandwidth maintenance traffic with a sufficient large field. We also give an evaluation to compare MCR with other redundancy recovery schemes in the storage cost and maintenance traffic to show the advantage of MCR.