I'm not done reading the article yet, but nothing so far indicates that this is zero-copy, just a more efficient internal representation

Nope. This is just a different implementation that greatly improves the speed in various ways.