I think this proposal is a little too vague; the premise seems to be based on the notion that a huge amount of specialized Wikimedia Foundation work is happening in core, when I find the opposite seems to be the case. Nearly all Wikimedia-specific feature development is in extensions (Cite, CentralAuth, WikiEditor, UploadWizard, ArticleFeedback, TimedMediaHandler, SocialProfile, etc), while core sees more infrastructure work to support them (eg ResourceLoader). Even highly useful general things are often done as extensions because it's preferable to develop and maintain them as modules: ParserFunctions, Gadgets, WikiEditor, a few bits in Vector, and all sorts of admin-y bits like CheckUser, RenameUser, etc.
I support the general goal of doing most feature-specific work in extensions, but that's more of a continuance of current practice than a change to it.
As we start bundling extensions with the installer soonish (probably for 1.19?) it'll also become easier to break things out that have been in core for a long time but are pretty standalone (lots of those misc reporting special pages!) or have other weird dependencies (eg Math which was finally broken out for 1.18).
If the suggestion is "keep at it", then I guess I agree; but this should probably be broken into smaller, more actionable goals:
- bundle some specific extensions with tarball for 1.19
- make sure installer can select some of those extensions by default
- make sure all those extensions actually work
- enumerate specific things to break out and start on them
Some things may require more work, such as figuring out how localization and dependencies mix in. For instance lots of people like to cite LanguageConverter as 'weird scary code' that should be broken out... but it's a dependency for several of our localizations which extend the base class and provide conversion data tables as part of the localization.
None of these things really has a deadline on it, and most of them are going to be highly independent of each other.