This paper propose a novel scheduling algorithm for Web switch operating at layer-7 of the OSI protocol stack, considering heterogeneity server performance. Its goal is to improve load sharing in Web cluster. The new policy classifies the client requests on the basis of their expected impact on server resources, and each request is given a special weight to compute the server load state. In order to lighten the computing workload of switcher, we improve the proposed scheduling algorithm, avoid to gathering load information of each server when dispatching request. The experiments show that the new policy can effectively balance the load of Web cluster system, and achieve more substantial performance improvement, especially for heterogeneous web cluster and highly dynamic content Web sites.