I'm doing an experiment with my recent papers: each has a Trac instance, so that code and the article text (a LaTeX file) can be shared between colaborators using Subversion. Data is hosted on an Opendap server, running my trusty pydap. But the best part is that the whole process, from downloading data to composing the article pdf, is done through a Makefile. The Makefile creates a virtual environment, downloads the required Python modules, and bootstraps the analyses. Just type make and enjoy.
Now imagine if all scientists embraced the open-source philosophy of open collaboration and real meritocracy and did the same.