Learning is conventionally thought of as something that happens "in someone's head". In this view collaborative learning is just an (unusual) technique for getting the knowledge into individual learners, and assessment should focus on evaluating how much knowledge each learner gained from the collaborative learning exercise.
It is important to realise, however, that learning is also a social achievement, and its traces can be seen in increased capacities of groups/ organisations/ institutions to collectively deal with problems, create knowledge and so on.
We aren't sure yet what types of assessment would be appropriate for such group learning, but one obvious possibility is program evaluation. There is a very large and theoretically (and methodologically) sophisticated literature on program evaluation, which could probably be made applicable here.
Empirically evaluating the effectiveness/efficiency of collaborative learning
In an article (Public Policy, Research and Online Learning) in Ubiquity, Stephen Downes (2003) argues that the effectiveness and efficiency of e-learning cannot be evaluated using traditional quasi-experimental designs as "not only does it create a new methodology, it creates new -- and unmeasurable, by the traditional process -- outcomes". The sorts of learning that are mediated by computers do not fit neatly into classroom situations and have become so pervasive that it would be an impossible task to tease out and hold constant particular variables.
Journal of Workplace Learning special issue
Volume 15 No 6 (2003) of the Journal of Workplace Learning (published by Emerald journals, unfortunately requiring a subscription) was devoted to the question: "Do networks learn?" Intro article: Tauno Kekašle; Riita Viitala (2003). Do networks learn? Journal of Workplace Learning, 15(6), 245-247. The intro article to the special edition. Points out that traditionally three levels of organisational learning have been identified - individual, work group and company-wide. They wish to in addition also consider the learning that might happen in inter-company networks.