Many modern computer and communication systems result in large, complex performance models. The compositional approach offered by stochastic process algebra constructs a model from submodels which are smaller and more easily understood. This gives the model a clear component-based structure. In this paper we present cases when this structure may be used to inform the solution of the model, leading to an efficient solution based on a decomposition of the underlying Markov process. The decomposition which we consider is time scale decomposition, based on Courtois's near complete decomposability. This work has been influenced by related work on stochastic Petri nets: we will discuss the advantages and disadvantages of taking such an approach to the development of techniques for stochastic process algebras. Our technique is illustrated by an example based on a closed network of queues with finite capacity in which blocking may occur.