As Erik points out, at a certain point we have to actually write new code to support new ideas. Else "projects we could do at Wikimedia" becomes "projects we can do with a wiki engine."
e.g. OpenStreetMap would have been a natural for WMF, but it would have required a whole new software infrastructure. And we have no shortage of content editors, but developers appear rather rarer.
Proposals I recall seeing for new projects either fit into a current project (e.g. Wikibooks - really, Wikipedia is a book, too) or haven't been neutral (e.g. the victims of Soviet repression proposal, which I think is a great idea but also think just would have been way too intrinsically non-neutral for WMF; the reviews wiki). Any proposal that's "hey, let's start a wiki" will, I suspect, fall into one of those two.
We're either not thinking outside the box enough or need to build new boxes. Or both.
What interesting new engines are there out there for gathering content from masses of Internet users that aren't wikis as we know them? What could we use them for besides their original purpose?
[cc'd to wikitech-l for comment as well]
- d.