The genetic algorithm implementation in IML is really nice, but the fact that it's a relatively new addition is obvious from what seem to me like a number of missing features.
The ability to change individuals within a population at any point. For example, I would like look at a subset of a population and make specific changes to their chromosomes. The workaround I came up with was to use gagetmem to read the current population, make the changes, and write a new initialization routine that forced the altered population to be read in again. This is obviously not a very elegant solution, but it works.
The ability to do preselection, "crowding", or "niching" to maintain genetic diversity through generations.