Another factor is that as the decades have gone by, in many branches of math the amount one has had to learn to get going researchwise has increased markedly. This of course varies a lot from subject to subject, but in many areas it can take quite a while after the PhD to develop the kind of research record that a research university wants to see before they hire you on the tenure-track and take the risks associated with that. This encourages the proliferation of postdocs as well.