Currently depth peeling assigns geometry into the last layer it can which does not cause an overlap error. This tends to have the first geometry evaluated (closest to the camera) put into the highest layers, and then most other things assigned to lowest layer.
A better solution would be to pack transparent geometry into the layers in a way which minimises screen space error. Doing this perfectly sounds a lot like the bin-packing problem!
Currently depth peeling assigns geometry into the last layer it can which does not cause an overlap error. This tends to have the first geometry evaluated (closest to the camera) put into the highest layers, and then most other things assigned to lowest layer.
A better solution would be to pack transparent geometry into the layers in a way which minimises screen space error. Doing this perfectly sounds a lot like the bin-packing problem!