50 - 60 fps are not good numbers based on the amount of units on screen. Also a lot of other factors go in to judge those numbers.
Polycount, bone count, number of skeletal blends per character/animation, character to character avoidance, obstacle avoidance, moving obstacle avoidance etc…
After revisiting this topic in a more professional manner I can safely say unless you use a third party tech, or a custom built one from scratch, getting any large crowd by the many hundreds and thousands to behave like proper looking crowd inside the engine is near impossible to achieve in its current state and byt he time of writing this.