The Synchronous Dataflow (SDF) model of computation by Lee and Messerschmitt has become popular for modeling concurrent applications on a multiprocessor platform. It is used to obtain a guaranteed, predictable performance. The model, on the other hand, is quite restrictive in its expressivity, making it less applicable to many modern, more dynamic applications. A common technique to deal with dynamic behavior is to consider different scenarios in separation. This analysis is, however, currently limited mainly to sequential applications. In this article, we present a new analysis approach that allows analysis of synchronous dataflow models across different scenarios of operation. The dataflow graphs corresponding to the different scenarios can be completely different. Execution times, consumption and production rates and the structure of the SDF may change. Our technique allows to derive or prove worst-case performance guarantees of the resulting model and as such extends the model-driven approach to designing predictable systems to significantly more dynamic applications and platforms. The approach is illustrated with three MP3 and MPEG-4 related case studies.