Because it's boring. And no browser implements it. You're not doing anybody any favors by implementing it. That google rich data thing? It's done with html classes too, that works equally well, and we've had them for a while now.
Seriously, I like microdata, I think it's useful, but this kind of utopian, far-away 'we should use this because it might be better in the future' never got us anywhere. That is the lesson we should learn from XHTML2, not 'adding more attributes is cool'.
I strongly disagree that it's boring and doesn't do anyone any favors. There are all sorts of domain-specific metadata that could serve as a platform for other services. For example, Creative Commons licenses parse RDFa in referring documents to render copy-paste attribution information (http://wiki.creativecommons.org/RDFa#CC_Deeds_using_RDFa_and...).
Standardized metadata can serve as a platform for other services, not just functionality in a browser. And the common sentiment that "we should use this because it might be better in the future" is a hurdle most standards go through at some point.
In any case, it's somewhat shortsighted to look for immediate, personal benefits. Of course you need to look at the payoffs for implementing a spec, but I think it's wrong not to take longer-term possibilities into account.
Well, as Cray mentioned, there are a variety of places where someone might be tempted to make a json or xml based API, and we can start implementing that as plain HTML5, and writing Python and Ruby libraries to make it as easy to work with as json or xml. That's easy, it can be done today, and it will lay the groundwork for browsers to actually implement.
Someday, it would be nice if every single street address on the web was marked up in such a way that just from looking at the markup, a computer can unambiguously figure out the location you're looking for. Or a way for browsers to unambiguously know that something is a phone number and not a zip code (Android does a horrible job of this. Try texting your current location from the maps app.)
I guess, the semantic web remains a dream, but that's no reason to avoid adding semantic data, especially if you want computers to interact with your data directly.
I agree that there's too much "utopian" technologies out there with impractical implementations. However, Microdata is already seeing Google support. Browser support I think could come, but we have to bring fuel to the microdata conversation rather than waving it off as nonsense.
Seriously, I like microdata, I think it's useful, but this kind of utopian, far-away 'we should use this because it might be better in the future' never got us anywhere. That is the lesson we should learn from XHTML2, not 'adding more attributes is cool'.