This paper proposes a methodological approach for the performance analysis of Web-based searching applications on the Internet. It specifically investigates the behavior of the client/server (C/S), remote-evaluation (REV) and mobile-agent (MA) communication paradigms and describes how Petri-net models can be developed to derive performance indices which can help the designer to improve the efficiency of his distributed applications. Our purpose is that of identifying a set of models that can help to understand the environmental situations in which such paradigms should be preferred or combined in order to optimize the performances of a distributed system. In particular, we propose a modeling technique applied to an information retrieval application on the World Wide Web. An analytical evaluation through the solution of nonMarkovian Petri-net models is provided, which allows us to identify the main parameters, as well as the way they interact, to be taken into consideration when distributed applications are to be designed. An experimental environment is also studied in order to obtain real measurements used to validate the analytical models.