I've weighed in a few times on how to build online communities or support scientists online, but it's really worth paying attention to when you get an actual scientist who is also very involved in and interested in social software tell you what he thinks. Cameron Neylon did just that in a recent blog post (comments on ff). I'll quote liberally from his blog and feedback some ideas.
(he uses SS4S to stand for social software for science) All of the numbered paragraphs are direct quotes from his post.
1. SS4S will promote engagement with online scientific objects and through this encourage and provide paths to those with enthusiasm but insufficient expertise to gain sufficient expertise to contribute effectively (see e.g. Galaxy Zoo). This includes but is certainly not limited to collaborations between professional scientists. These are merely a special case of the general.
There are a couple of interesting thing there - first that "citizen scientists" and interested non-scientists are welcome and encouraged to participate in the same tool. They are provided support to move from legitimate peripheral participants  to more central contributors. So contrast this with the concern about "the public" seeing how the sausage is made. I found in a study I did for a class a few years ago that the quickest way to kill an online community of engineers was to have undergrads inundate them demanding homework help.
On the other hand, many scientists do want to engage with the public for lots of reasons, so supporting that is a good thing. This all feeds into some stuff I've been thinking about recently about how to sort of merge scholarly communication models with popular communication since your communication venues are findable and useable by the public (I like a lot of what Meyer and Schroeder say )
2. SS4S will measure and reward positive contributions, including constructive criticism and disagreement (Stack overflow vs YouTube comments). Ideally such measures will value quality of contribution rather than opinion, allowing disagreement to be both supported when required and resolved when appropriate.
Good policies , good moderators, charitable reading, a way to comment on the specific thing you mean and to do so clearly.
3. SS4S will provide single click through access to available online scientific objects and make it easy to bring references to those objects into the user's personal space or stream (see e.g. Friendfeed "Like" button)
Absolutely. And then be able to interact with these things, annotate them, and then re-mix them into other things.
4. SS4S should provide zero effort upload paths to make scientific objects available online while simultaneously assuring users that this upload and the objects are always under their control. This will mean in many cases that what is being pushed to the SS4S system is a reference not the object itself, but will sometimes be the object to provide ease of use. The distinction will ideally be invisible to the user in practice barring some initial setup (see e.g. use of Posterous as a marshalling yard).
5. SS4S will make it easy for users to connect with other users and build networks based on a shared interest in specific research objects (Friendfeed again).
What metadata is required for this? What data must the system store and use to make this work?
6. SS4S will help the user exploit that network to collaboratively filter objects of interest to them and of importance to their work. These objects might be results, datasets, ideas, or people.
Or models or equations or modules...
7. SS4S will integrate with the user's existing tools and workflow and enable them to gradually adopt more effective or efficient tools without requiring any severe breaks (see Mendeley/Citeulike/Zotero/Papers and DropBox)
8. SS4S will work reliably and stably with high performance and low latency.
9. SS4S will come to where the researcher is working both with respect to new software and also unusual locations and situations requiring mobile, location sensitive, and overlay technologies (Layar, Greasemonkey, voice/gesture recognition - the latter largely prompted by a conversation I had with Peter Murray-Rust some months ago).
That's pretty cool. I mean mobile is fairly common, and location sensitive things are not uncommon, but these with overlay (like augmented reality? hmm, that could be very useful for sharing protocols)
10. SS4S will be trusted and reliable with a strong community belief in its long term stability. No single organization holds or probably even can hold this trust so solutions will almost certainly need to be federated, open source, and supported by an active development community.
Stability, reliability, and clear policies and provisions for preservation are important.
If you have comments on any of these or other suggestions, please leave them on Cameron's post or on friendfeed (or on here, I'll pass them along).
 Lave, J., & Wenger, E. (1991). Situated learning : legitimate peripheral participation. New York: Cambridge University Press.
 Meyer, E. T., & Schroeder, R. (2009). The world wide web of research and access to knowledge. Knowledge Management Research and Practice, 7(3), 218-233. doi:10.1057/kmrp.2009.13
 Preece, J. (2000). Online communities : designing usability, supporting sociability. New York: Wiley.