See this blog post on version control co-authored by Titus Brown:
Motivated by much the same problems as programmers – how do I track changes and authorship, and resolve conflicting changes? – scientists have increasingly begun using these version control systems not only for software but for paper and grant writing, and even for tracking data and metadata. In particular, CTB and others have argued that version control should be a required “good practice” for all computational science in practice. This increased interest in versioning for science has led to a realization that, collectively, we are lacking in workflows and practices that take advantage of version control. This is especially true for git and Mercurial, both decentralized version control systems that enable many different modes of collaboration.
I especially like the emphasis on how version control can facilitate researchers using others’ output and work more readily:
One of the most interesting potential uses for git and Mercurial is in supporting the “forking”, or copying and modification, of data analysis pipelines, to enable field-wide reuse of the analysis pipeline provided in a published paper. Unlike more centralized systems such as Subversion, git and Mercurial both readily support diverging development of software starting from a common base by other groups. This kind of branching enables remixing of research software, while retaining all provenance information. Because projects can easily be made available for branching upon publication by using GitHub and BitBucket for post-publication development and dissemination, many barriers to remixing and reuse of software are significantly lowered.