I will return tomorrow with a more detailed response, but unfortunately I have to leave now.
I made an attempt at replicating the SC2 movement behavior and this is what I managed to achieve in 30 minutes or so (actually this is not the latest result, it looks slightly better now):
It uses local avoidance as well as some custom logic for determining when agents should stop (to avoid them all trying to constantly move towards the same point). I can post the code if you want (it’s about 70 lines which will work with the latest 4.1.2 beta).