The main reason is **inertia and lack of information**, I think.

* Researchers are not really aware of the costs their institutions have to face to subscribe to journals. From their perspective, publishing and reading are mostly free. It is not always easy to tell whether you have access to a PDF file because of your university's subscriptions or because it is open access. As long as they can download the file, they are happy, regardless of its origin. So, when they publish articles, they sign the copyright transfer agreement form without reading it, as they do not consider the publishing industry as a problem.

* Universities are unaware of the existence of rights retention policies (or even open access policies, in fact), or do not take the time to adopt them because they fear it will incur an unpopular burden on the faculty. Moreover, implementing such policies is complicated, as it requires an institutional repository and a research information system keeping track of all the publications, their compliance with various policies. Shameless plug: we are building [dissemin](http://dissem.in) to solve that problem.

* Publishers leverage this inertia and laziness to secure a steady stream of income. They lobby to convince stakeholders that open access necessarily implies high charges at publication time. They also polish their image, by sponsoring conferences and learned societies. They give free subscriptions to Wikipedia editors and Article Processing Charges waivers in poor countries.

To answer the OP's comment: if you, as a researcher, simply release your paper in CC-BY (say) before signing the copyright transfer agreement form, then you consciously sign a form granting exclusive rights you cannot grant anymore: it does not work. That's written in section 205(e) (emphasis added):

> (2) the license was taken in good faith before recordation of the transfer and **without notice of it.**

That is why you need an institutional policy granting these rights automatically for the researcher. For instance, [the University of California has contacted as many publishers they could][1] to give them formal notice of their policy, which allows researchers to sign the CTA forms without amending them.


  [1]: http://osc.universityofcalifornia.edu/open-access-policy/publisher-communications/