Packet losses can significantly degrade the quality of service (QoS) provided by networks. In this paper, asynchronous bufferless optical packet switched networks with multiple service classes and full wavelength conversion are considered. In case of a large number of wavelengths per fibre, packet losses become rare and thus intractable by means of conventional performance methodologies. We present a novel fast simulation technique for efficiently estimating the loss rates in this setting. Our technique applies Importance Sampling which proceeds by using an alternative probability distribution for accelerated simulation and appropriately weighting the results to obtain unbiased estimators. With our technique the necessary parameters determining the alternative distribution are adaptively obtained by ant colony optimization, a metaheuristic that is in widespread use for solving combinatorial optimization problems. Thereby, no intimate a priori knowledge of the system under consideration is required to set up the simulation. Rather, the parameters adapt to the model. The accuracy and the efficiency of the novel technique are demonstrated by numerical results.