One and a half year after creating XFML, I think the main value of the language has been to introduce people to concepts such as facets and topics.
It has become the de-facto standard for exchanging faceted metadata, but on the other hand, very few applications are actually exchanging that kind of information. I think that'll change, but that's a separate story.
It's simplicity has held up well, both conceptually as implementation-wise. Most people understand the spec easily, and it's easy to implement.
When creating the spec, I spend months removing elements and attributes that I thought weren't absolutely necessary. There are a few in there that nobody's using, like the connect element, but I do have some hope they may still be used and become important one day. Then again, maybe I should have made the langauge even simpler. Extreme standard creation: don't include elements you're not entirely sure you'll need.
Another element I'm not sure about is MapInfo. I'm curious to see wether that one will be really used one day.