This is really a blog question, with no single answer.   There are 150+
Ph.D. programs in mathematics in the U.S., plus various institutes.  As
noted already, there is no central planner.   Before the 1960s there
were relatively few mathematicians in the U.S.   Those doing active
research in universities did a lot of routine teaching and were low paid.
But then higher education mushroomed for many reasons, with new programs
and universities all over the place.   The Ph.D. program here at UMass
(earlier Mass. Agricultural College) started in the mid-1960s.   Hiring
everywhere of new Ph.D.'s was frenetic and worked mainly through the old
boy network.   As the market settled down and recessions occurred, 
temporary postdoc positions became more common.   Some at the prestigious
universities were institutionalized, others were an afterthought and most
often on a 1-2 year basis.   At UMass almost everyone in the 1970s was
on a tenure-track, with just a few non-Ph.D. "lecturers" on contracts.
It's cheaper for administrators to hire non-Ph.D. people just for teaching
service courses, so here it was difficult to persuade the provost and others
that postdocs are standard for research-based math programs.  Our main
postdoc burst occurred after two state early retirement incentives removed
most of the senior faculty.   Now postdocs are sporadic and hard to authorize
here, though the department still wants them.   (There are still a few
lecturers, who have full faculty benefits but no tenure.   We'd have more
if they didn't undercut faculty teaching ratios and if UMass had built
bigger lecture halls.)   The statistics part of the UMass department still
tends to hire new Ph.D.'s on tenure tracks due to market pressures from
nonacademic employers.  No conspiracies, but lots of pressures on the system.