The optimal assignment system is single task based, where a single task is carrying a single-resource. Basically, we develop exactly what a unit is going to do, carry a single resource (or other possible tasks), and calculate the exact distance it will cover to do that task. If its carrying a resource, this is just distance to the building, if its not, it calculates distance to resource, and from that spot, distance to building. Its important that the algorithms and gradients we use here are the same as the ones the unit will actually use to determine its path. Time is the only factor affecting efficiency, and distance is the major factor affecting time (we could take walking/swimming speed, and even harvest speed into account if we wanted to).
To be truly optimal: We then consider all buildings, all building unit combinations, and find the one that fills all possible slots that ends up with the lowest overall distance covered. This could be run as often as wanted to take care of changing circumstances.
-- Extra cheese comes at a cost. Bradley Arsenault.