I'm happy that this idea has found a good environment to flourish in. It's going to be a lot of fun and you're going to improve a lot of people's lives as a consequence of it. Enjoy it, and keep up the excellent work.
Databases make the most sense when your data changes frequently but if you have something like congressional districts, census information, etc. where the data changes infrequently but is likely to be shared widely the DVCS workflow and simplicity are really appealing. Right now you could do quite a lot with e.g. a gh-pages branch which contains GeoJSON and a simple LeafletJS viewer to combine files into layers; being able to quickly preview data files just makes that easier and it seems unlikely that they won't be adding some sort of diff comparison feature in the future.
Will give you all of the results, it definitely needs a slightly better way of searching.
Consider the last time there was an election. Those dynamically updated maps use this kind of file.
And without this support, shape files can't be viewed at all in a GitHub project, which is actually kind of ridiculous in 2013.
Furthermore, it encourages shape file providers to move their content to GitHub so we don't have the standard scenario of "download this .zip file and compile it to your preferred format to see if it's worth your time". It streamlines the process significantly.
It also gives them an edge over BitBucket. :)