There are two major types of open access archives, institutional and disciplinary (or subject) archives. This post explains why disciplinary / subject archives are optimum for searching and have an advantage in the short term for growth, as well as why institutional archives are also necessary.
For the searcher, the optimum will almost always be the disciplinary / subject archives. In the short term, this gives a tremendous advantage to disciplinary archives like PubMedCentral, arXiv, rePEc,and E-LIS. Researchers generally tend to search for literature primarily within their own discipline, so this approach makes sense to the researcher. arXiv is heavily used by physicists, so of course physicists want to deposit their work here, so that it will be read. It is no wonder, then, that the disciplinary archives are relatively successful. PubMedCentral is the world's largest open access archive, with over a million items.
Medical research funding agency policies require deposit in PubMedCentral. This makes sense. Literature that is deposited in PMC is more useful for searchers. This locus of deposit fits with the preference for searching medical literature. This creates incentives towards open access beyond the literature mandated for open access. We are already seeing this with hundreds of journals voluntarily contributing all contents to PMC, many for immediate open access.
Institutional archives are also necessary, for a number of reasons, two of which are listed here. First, the majority of disciplines do not have disciplinary archives, so this is necessary. More will likely be developed, but given that many institutions now have archives, future disciplinary archives may well be developed within institutional archives. Second, institutional archives will provide functions and services to institutions beyond what disciplinary archives can do, such as showcasing the work of the university, its departments and faculty to potential donors, including the public and politicians for publicly funded institution, and potential students. The more substantial the research output of a university, the more the university stands to gain from the institutional repository.
Software for open access archives is being developed to facilitate cross-deposit, so it does not disadvantage one type of archive if an open access policy requires deposit in another archive.
Terminology note: the author uses the term "open access archive" where others use "repository". In my opinion, "open access archive" clearly states the purpose of the service, i.e. providing open access and preservation, in a way that "repository" does not. It is understood that not all items in such archives will be available for open access.