Quantifying the performance of component-based or service-oriented systems is a complex task, e.g., it is non-trivial to calculate the end-to-end quality of service of a composite Web service. An established approach to reason about such systems in general is the use of coordination models, which can provide a formal basis for both their verification and implementation. An example of such a model is the channel-based coordination language Reo and its probabilistic extension Stochastic Reo. However, all existing performance analysis approaches for Stochastic Reo are restricted to the use of exponential distributions. To this end we introduce a transition structure, which enables a simulation approach for performance evaluation in Reo, enabling the use of arbitrary distributions and predefined probabilistic behaviors. Our approach supports steady-state and transient analysis and, moreover, scales much better than the existing automata-based algorithms.