Maybe I'm missing something. Are you saying that in your application, handling rarely-occuring unanticipated disconnections via exceptions has such high overhead that its results in unacceptable performance?