In this paper we develop a method for analysing and comparing the performance of different testing techniques for concurrent systems, and use it to give some evidence that the so-called "exploration testing" finds errors faster than traditional testing based on test cases. We model the system under test as a state space with a weight and cost assigned to each transition, and find the probability and expected cost of reaching terminal states. From this information, the probabilities and expected costs of finding errors using each method can be computed. A drawback of our method is that it is not feasible for arbitrarily large systems, but, in return, it gives results much quicker and with much higher precision than possible by running actual tests.