The most common problem in multipacting computations is not using a fine enough mesh, particularly in the areas where multipacting may be occurring. Multipacting resonances can be very sensitive to the mesh until a sufficient mesh density is achieved. Since there is no adaptive mesh refinement support for particle simulations, it is a good idea to try different mesh densities to ensure that you have a well-resolved result.
Another common problem is to base the field level used in the particle simulation on a quantity that is highly mesh sensitive. For example, if the structure has a conducting corner the electric field there is technically infinite, and will in practice become increasing large as the mesh is refined. In this circumstance choosing the field level normalization target to be Peak Surface |E| can lead to highly variable results. In situations like this it is better to choose a normalization target that is not so mesh dependent, such as Peak Axial |E| on Axis, or Average Accelerating E-Field Magnitude.
Not sourcing enough particles can lead to missed resonances. Since the particle sourcing is a random process, it is important to have a sufficient number of initial particles to ensure that at least some of them have the correct velocity and position to initiate a resonant orbit. Sourcing more particles will make the problem run longer, so the best approach is usually to start with a relatively low number and then run a case with larger number to validate the previous result.