I think the basic question should be different. Why do you want to find out if there is a population structure? What is the underlying scientific question? This will determine which tests you will need to use, and therefore which markers to choose and how many, not the other way round.
As you try to infer genetic structure, or genetic distances, ask yourself why you are doing so? If you get result X, what will your interpretation be? And what if you get result Y? What can you deduce about the four major processes (drift-Ne, selection, migration, mutation) that affect the genetic structure of your populations? If you want to know a pattern, you must have a prior idea of what processes shaped these patterns.
As for the dogm that you should use stepwise mutation models for your inference of population structure with microsatellites, the same reasoning goes! If you are interested in phylogeographic aspects of your populations, in long-term averages of estimates of population parameters, use such mutation models. Use them if you think that mutation contributed to the pattern you are investigating, and if it is relevant at the time scale at which you are making inferences about the underlying processes. If however you are interested in the recent processes affecting your populations, stay away from them! If you want to know about the recent aspects, taking the evolutionary history of alleles into account is not a good idea. Mutation events that may have happened thousands of years ago and thousands of kilometers away in now extinct populations, but that nevertheless led to alleles that are apparent in your current populations will add noise, not signal. Admixture of populations is especially problematic, as two processes may be at work here: migration and mutation, and using explicit mutation models, you assume that migration did not contribute to the observed genetic structure (allele 120 and 122 may have come together by admixture in your population, but you assume a mutation event caused that. Yes a mutation event did cause it eventually, but not in your focal population, but somewhere else and maybe a long time ago).
Basically, this typically comes down to violations of underlying assumptions of a test. Know the assumptions of the test you are using. For example a bottleneck test will test for a deviation of mutation-drift equilibrium, and (often silently) assume that migration is negligible. Using a stepwise mutation model in a population that was likely affected since its creation by migration to a similar (or larger) degree than mutation will lead to erroneous conclusions about the deviation of mutation-drift equilibrium. If, on the other hand, you use an infinite allele model in such a case, you make no inferences on what caused an eventual deviation from equilibrium: mutation, migration or both. So depending on your questions and underlying assumptions, you should use explicit mutation models (stepwise and so on) or implicit (infinite, k-allele) mutation models.
There is no single best mutation model: the marker does not define the mutation model to be used, only the scientific question does (and thereby it determines which marker type and how many you need).
A late answer, but I hope it 's still useful.
Kingsly Chuo Beng
Chinese Academy of Sciences
Wildlife Institute of India
University of Belgrade
Chiang Mai University
Institute of Oceanography and Fisheries
Central Marine Fisheries Research Institute
Anthony J. Giordano
Jonathan Lyon Whitney
University of Hawaiʻi at Mānoa