yes, it is not optimal at all.
There could be a solution by using an invisible group leader actor (should be pawn or character without collision), that does the pathfinding and movement along the long path to be followed by units.
Group member units should update their target within a couple of frames to decrease performance cost, and to improve obstacle avoidance, while keeping more or less the formation shape (formation facing direction also can be updated by setting similar to the invisible actor’s direction, to calculate the offsets of unit target positions).
If units lag behind a lot because of e.g. collisions with another group, the invisible actor should slow down or stop until units get closer.
Moreover, groups could avoid each other if the invisible leaders could manage it e.g. by setting their size as large as group width (they could collide with each other), and some movement prediction could also help.
I think it would not be very complicated to implement, and must work much better than my example described above 