The second kind is stress testing in its original form. The application is encountered with a peak volume of data over a short span of time. So if the just mentioned air-traffic-control-system was designed to monitor and guide two hundred planes, simulate the occurance of those two hundred planes. Do not hesitate to confront it with an unexpected 201st plane, and check its response time on such an event.
Although it is not unlikely to construct situations this way that may -- if at all -- only rarely occur, those tests still have their justification. First of all, it is not possible to lock out those situations, and so especially in real time systems, they have to be tested. Just imagine the facial impression of the operator in the tower if the system suddenly asks: ``There is a 201st plane. What should I do?'' or if it even presents him an error message. Secondly, errors might occur that also happen in less stressful situations with a lower probability.