Long-overdue and laudable. Interesting to see how it plays out and whether other fields can follow suit; I think my own would take to this idea fairly well.

Also interesting to see that they haven't decided to change much about the peer review process. The arXiv overlay structure prohibits a practise that I've been hoping to see introduced, which is blinding reviewers to the name/s and institution/s of the author/s.

Does "double blind" work in your field? It does not in mine (compilers, programming languages).

As an author, I can usually identify the anonymous reviewers as well, if they are not outsiders.

The vast majority of papers in quantum information and foundations involve proofs of some mathematical results and some discussion of their physical/operational relevance. I don't see anything to prevent it in principle. What stops it working in your field?

The community is small and everybody knows each other.

There are several arxiv overlay journals. Example: http://www.emis.de/journals/SIGMA/about.html