Model checking of probabilistic models can be done either by numerical analysis or by simulation and statistical methods. In this paper, we compare the efficiency and the scalability of these approaches when they are applied to the verification of steady state properties of very large models. We provide an experimental comparison study between the statistical model checking using perfect sampling, the numerical method implemented in model checker PRISM and the statistical model checking implemented in model checker MRMC for the verification of CSL steady state properties. We show that the statistical approach using perfect sampling is generally more efficient than the two other approaches and it allows us to consider very large models and to verify rare event properties efficiently.