Seemingly overnight, the Internet has gone from an academic experiment to a worldwide information matrix. How to improve the response time and throughput of a complicated network system is a challenging problem. Discussed and provided in this paper are quest dispatching and selecting schemes for Web server clusters and the stochastic high level Petrinet models for those schemes. A novel method is proposed to allocate Web request tasks according to the minimums finished time and an analysis is also presented based on the experiments using Petri net. Experiment results show that the method is feasible and effective.