I am a federally funded researcher who administrates a program that publishes quite a bit. First off, I am a supporter of open access publishing. Here is our challenge with the current policy, and why it has been very difficult to adopt.
Open access journals cost between $1-3k per publication (see PLOS or BMC). These journals automatically submit papers to the public repository. This is a direct cost that comes out of my grants that may not have been originally budgeted. Now, closed access journals are generally free or close to free to publish. The new policy requires submission of closed access papers, by the authors, to the central repository (if federally funded). Obviously, this violates the agreement the author had with the publisher, so the author, on their own, must negotiate a legal mechanism to do this. Some publishers charge to do this, maybe more than $1k. Every submitted paper gets an ID that must be submitted with a progress report. When we publish 5-10 papers per progress report, this is frankly a lot of work and sometimes, we fund papers partially that are published by other groups. So it is up to me to encourage these groups to figure this out, so I can include them in my reports. More work, and it adds another level of complexity to collaboration.
So far, this has been an administrative headache, it is expensive and considering most major university libraries already have licenses to the closed data, it seems, to me, unnecessarily complicated. I wish they had required the publishers to do this (each publisher would have to work with one source) instead of the researcher, because we have to work with a number of publishers and that takes time in an already very, very competitive field.
There are some really great aspects of open access publishing and the power of the resulting knowledgebase of manuscripts is going to be really exciting, however, $10-20k/year for page charges is only going to result in less science, IMO.