In order to circumvent the problem of state-space explosion of large-scale Markovian models, the stochastic process algebra PEPA has been given a fluid semantics based on ordinary differential equations, treating all entities as continuous. However, low numbers of instances and/or relatively slow dynamics may make such approximation too coarse for some parts of the system. To deal with such situations, we propose an hybrid semantics lying between these two extremes, treating parts of the system as discrete and stochastic and others as continuous and deterministic. The underlying mathematical object for the quantitative evaluation is a stochastic hybrid automaton. A case study of a client/server system with breakdowns and repairs is used to discuss the accuracy and the cost of this hybrid analysis.