Using this discussion post as a location to help brainstorm ideas for simulating crowds. The current problem is that CryAnimation always wants to start animations at the same exact frames, making this difficult to create a true crowd simulation.
One option could be to do multiple Mannequin setups and offsetting the animations within the Fragments. In the Anim Clip Properties for the animation within a Fragment, apply some value to the Start Time property to have it start on a different frame. The benefit of having multiple Mannequin setups is that you could then create multiple character slices that reference one of the Mannequin setups to make it easier to instance the character variants for your crowd. Using the slices should be more performant as well in the level.
Alternatively, if you’d rather only have the one Mannequin setup to deal with, you could set up multiple similar Fragments using the same animation but with the Start Time offsets applied. The trick will be getting a character to map to one of the offset Fragments instead of defaulting to the first entry, which may have to be forcing specific characters to always have a specific tag on at all times to always use the Fragments offset with that tag.
I would recommend the first option. I think it will be cleaner to implement even though you will have multiple Mannequin setups to maintain, but the referencing will ensure that it always uses those specific offset animations in that Mannequin file instead of trying to use alternative Fragments. This way you can have X amount of character slices as well for your default variants based on animation offsets, but still be able to vary their appearances even further through the character skin/joint attachment system.