Multi-agent systems (MAS) are one of the complex applications of distributed artificial intelligence. They are prone to different kinds of exception due to their characteristic of operating in a complex and dynamic environment. The dynamism and unpredictable nature of an open environment gives rise to unpredictable exceptions. It becomes essential to have some exception diagnosis mechanisms in place to be able to diagnose the cause of such exceptions and to execute proper recovery plans. These mechanisms do come with some overheads. In this paper we present an empirical evaluation of our proposed sentinel based approach to exception diagnosis in an open MAS and also discuss the trade offs in using a sentinel based approach to exception diagnosis in an MAS.