According to the performance requirements of distributed applications, the load balancing is introduced into fault tolerance in this paper to reach tradeoff between active replication algorithm and primary backup algorithm. A dynamic fault tolerant algorithm RAWA (read-any-write-any) is presented in the paper for redundant servers to dynamically change quorum with variability of system load. It not only improves the processing speed of requests, but also implements load balancing in a more efficient and simplier way. Integrated with consistency and mutual mutual visiting mechanisms, RAWA algorithm can be applied to stateful servers and nested visting. The performance of RAWA algorithm is also analyzed, and through experiments on CORBA platform, the comparisons with other fault tolerant algorithms indicate that RAWA has effectively improved the performance of redundant servers without damaging system availability.