Repository as Platform? Or Product?

Is repository software (DSpace, EPrints, Fedora) a platform to build on, or a shrinkwrapped product to unpack and use? There are at least two answers to this question, and each software has to try and strike the right balance for its intended community.

I note the results of the recent DSpace community survey, that shows that 80% of repositories use the default metadata configuration, 78% have made at most "minor cosmetic" changes to the configuration and 62% use no addons beyond the distributed core code (stats, SWORD, google indexing etc).

This seems to support the view that if you come up with a new feature but it isn't a standard part of the core repository then it won't be used. It's a challenge for repository software designers, and for repository projects. For example, how do you make the repository user interface pleasing and useful to artists, engineers, teachers and researchers all at the same time.

Answers on a postcard, please!