Emily,
I would go ahead and estimate the migration rates, you can compare also the 5 population model (I suggest to use a model that allows fewer than 20 migration rates) and the model with 4 populations (the two populations combined),
I assume these model may have rather different migration parameters. I am relatively skeptical with any ABC analysis that has many parameters because usually not enough datasets are sampled and summary statistics are chosen haphazardly, to me ABC is equivalent with Monte Carlo (MC) integration. MC was replaced my Markov Chain MC in 1953 for complex problems. But philosophical issues aside, I find it difficult to believe that your 10 microsatellite loci can pinpoint the population merger that precisely: questions to ask the ABC method: mutation model? How are population simulated forward (almost impossible for large populations) or backwards?
If you have populations that are admixed then you will get most likely a large population size, but I am not clear whether this has a huge effect on the migration rate estimates.
Peter