Enabling scalability-sensitive speculative parallelization for FSM computations