Event-B is a tool-supported specification language that can be used e.g. for modelling of concurrent programs. This calls for code generation and a means of executing the resulting code. One approach is to preserve the original event-based nature of the model and use a run-time scheduler and message passing to execute the translated events on different computational nodes. In this paper, we consider the efficiency of such a solution when applied to a compute-intensive model. In order to mitigate overhead, we also use a method allowing computational nodes to repeat event execution without the involvement of the scheduler. To find out under what circumstances the approach performs most efficiently, we perform an empirical study with different parameters.