And then one comes up against the 800kg problem of authentication. Originally all this collaboration was a bit ad hoc - a research group needs a blog - install ubuntu in server mode on an old pc in a corner of the lab, install apache, php and wordpress. Create some local accounts, not forgetting one for Joe who'd moved to another institution, and hey presto, we were blogging. Spontaneous, effective and fun - well at least for those research groups that had an IT literate postgrad, who then moved on elsewhere and no one knew much about maintaining it.
Or take the other scenario, the middle aged lecturer in middle english who starts this really interesting blog on the divergence of Frisian English and Dutch around the time of the great vowel shift. Of course he doesn't know much about blog software so he gets a hosted wordpress or blogspot account and then invites a few of his mates to contribute posts. Very ad hoc, very spontaneous and totally invisible to any research quality assessment exercise.
But using small scale resources means that we don't need to have complex authentication, and because everyone's mates, everyone trusts each other.
The only problem is that these blogs increasingly represent a research output and a form of scholarly communication, which means that universities want to host them, if only to ensure their backed up. And hosting of course means hosting them in a multiuser environment with properly provisioned accounts, something that a lot of blog software isn't really designed for, hence projects like lyceum, wordpress-mu and their various ldap plugins.
And that's where it stops. Authentication only is within the institution's ldap domain, or in the case of some of the large system universities in the States, the particular college's ldap domain, effectively killing the spontaneity of collaboration stone dead.
Fortunately there is an answer - shibboleth, which has kind of been a wide area authentication mechanism without a purpose. Everyone thinks its a good idea, but really, until collaboration is required, it remains mere geekery. The joy of shibboleth is that while the implementation can be no easy exercise, for the end user it can be as easy as using openID to gain access to services. It does also require that the shared resources are shibbolized to work with a shibboleth basd mechanism. Given that many of the products required are open source this is less of a problem than it might be.
Also as shibboleth gives users control over what attributes are released the amount of information disclosed is consensual.
The only real problem remaining is a mechanism to provide access for people affiliated to institutions with no IdP, or visitors from outside of the academic world. There need to also be a mechanism for institutions to 'sponsor' non-affiliated invitees to get a non-affiliated shibboleth account that may be more restricted in scope but which will allow them to work with groups of affiliated researchers across institutions. Such a service could be provided on a per institution basis, or on some other basis, for example one provided by the government for employees engaged in collaborative research rather than on a per agency basis.
The mechanics don't actually matter, it can be one solution or a mix of solutions, the main problem is to ensure that whatever solution is found encourages openness and the free flow of communication both within institiutions and across institutions.