In this paper, we discuss temporal performance of supervised discrete event systems by comparing two supervisors for one discrete event system. We show that the temporal performance of the supervised systems can be inferred by their logical properties described by the languages generated. More precisely, for two supervised systems, if the language generated by the first system is contained in the language generated by the second and if the language generated by the second satisfies a certain fairness condition with respect to a set of events of particular interest, then every occurrence of events in the set occurs earlier in the second system. In this sense, the supervisor generating the larger language is better. We extend this result to the asymptotic behaviors, where the conditions for optimality are weaker. An example is given to illustrate applications of the results.