The diagrams in that way are problematic. When you model them as digraphs and use this layout and order them from too to bottom you get all the tables at the top that represent n:m relationships and at the bottom (at different levels) your main/independent entities.

At least bottom to top would pack the independent ones to the top and those that represent many-to-many relationships at the bottom.

From my work with Oracle Datamodeler know that even at a dozen tables manual layout beats an algorithm any time. And the oracle algithms are more sophisticated/less general than dot.