There's a reason why some bodies ship a reference implementation with the standard. Sometimes what's specified is infeasible, or even impossible (if the spec contradicts itself) to implement.
I think sometimes this is because the full spec is infeasible to implement, at least in a performant manner, and rather than implement a fully compliant but slow mess, they implement a partially compliant but usefully fast subset.
Providing a reference implementation allows a standards body to learn early that what they are speccing is counterproductive in one manner or another, prior to releasing it and getting vendor feedback.
Anyway, I totally agree that there should be a reference implementation. Imho, preferably in a purely functional language, to keep it as clean as possible.
Oh, I'm not defending Microsoft specifically, they're generally pretty horrible on this front. All the browser vendors have components in various states of compliance though, and some components seem to sit at a certain level for years.