Those constraints don't seem that strong to me. The algorithm could've just been retrained with different stack sizes if it ended up making a big enough difference.