When you say that any certification process (for a non-trivial field) necessarily becomes a sampling process, it is implied that this is already happening, i.e., there must be already a sampling process going on in the current universities. Decoupling may not solve this "problem" (if you consider this a problem), but it also does not create another one. It simply makes it more flexible so people who could not afford the education process for any reason, but were able to educate themselves by alternatives means, can also try to get certified in the level they wish. Also, the possibility of being certified by top tier certification emitters works as another motivator for people to strain and study harder (as going to top universities is not even remotely possible for most people, at least being certified by top certificate emitting institutes is probably less expensive (given the educational process is the more expensive one), and thus possible).

In other words, I believe you are right when you point that the educational institutes would narrow their focus to cover only the necessary bits to get certified, but I don't see this getting much far from what we have already; and from the point of view of the certificate emitting institute, it is desirable for the top ones to not emit certificates to anyone (in order to build a valuable brand name), having them to focus on elaborating challenging tests. Of course there may be "certificate mills", but again, this already happens, just with less flexibility.

What really picks me is that, at the end, it is all about the value people give to these certificates. Nothing would prevent society (including employers) to value more the educational process than the testing one; nothing would prevent the educational institutes to also emit certificates, thus coming back to where we are; but employers should not rely purely on certificates anyway, then I believe there is hope for the overall proposed model.