This is not compression. The author focuses on the small number of nodes and calls it compression but it's the edges that encode the information and there are a lot of them.
As for how well this works compared to other approaches, I don't know, and this blog post certainly doesn't address the question.