The use of Web Clusters that implement Quality of Service has gained much attention in recent years. In this paper we propose an adaptive algorithm for a Web Switch that estimates the best configuration of the Web cluster attending to the throughput of servers. The slotted time used by the system is adaptive to the burstiness factor of client requests arriving to the Web Cluster. The web switch model is based on the estimation of the throughput that servers would have in the next future to dynamically balance the workload. In order to reduce the checking time, estimations are computed in a variable slot scheduling. Simulation results are confronted with other algorithms and show that the algorithm proposed performs significantly better in terms of latency time for priority users.