I'd like to see the percentage improvement over the "naive" approach (just stacking them edge-to-edge) for each of them. I can't tell if the improvements are significant enough for this to have obviously practical benefits.
Obviously(?) the improvement decreases with increasing n. From the table in [0], regarding the length (not area) of the square, it is about 10% for n=5 and 7.5% for n=10, and 4.6% for n=82.
