It is this change in 3.2:

    randrange is more sophisticated about producing equally distributed
    values.  Formerly it used a style like ``int(random()*n)`` which  '
    could produce slightly uneven distributions.

-                return self._randbelow(istart)
+                if istart >= maxwidth:
+                    return self._randbelow(istart)
+                return int(self.random() * istart)

by rhettinger. Since there has not been any regression tests that the seeded numbers would stay compatible, they don't. Perhaps it would be a good idea to *add* such tests.