This paper deals with the definition and implementation of a distributed Web-based architecture for the on-line power system security analysis. The architecture employs field data measurements to dynamically estimate the actual state of an electrical grid and solves, for each possible contingency, the power system state equations to classify critical contingencies that could lead to system vulnerabilities. To identify the critical contingencies a computational engine able to execute the power system state equations is implemented according to a variant of the master/slave pattern. This pattern is defined to easily leverage the hierarchical network topology of resources connected to the Internet. The pattern is implemented with both RMI and ProActive, while the resulting framework is deployed on a network of heterogeneous clusters and used to compute the contingency analysis of a realistic electrical grid. The experimental results demonstrate that on-line security analysis can be executed in few minutes even for large network complexity.