Disclosure: I am of one of the developers of DEAP.
DEAP is the most actively developed project amongst the ones mentioned. It has an active mailing-list, which is an interesting feature if you need help at some point. The class creation which is unique to DEAP makes switching from single to multiple objectives really easy. It comes with multiple examples, including examples of multiobjective genetic algorithms.
It is also compatible with both Python 2 and 3, while some other frameworks only support Python 2. Finally, while it is written in pure Python, we will always have performances in mind, so it is quite fast. Timing of the different examples are available at http://deap.gel.ulaval.ca/speed/.