It's the difference between people who have had to troubleshoot obscure problems in complex systems and people who haven't.. Software archeology is important.

My personal route: clean atomic commits that do one thing at a time and are justifiable on their own, rebased to minimize small-scale concurrent branches littering the history.