I'm not so familiar with what we can do with UMAs, but I would be very interested in three metrics here, depending on how feasible they are:

1. Pages with any <template> instances at all
2. <template> in static HTML
3. <template> created dynamically
4. Cloned template contents via document.importNode(template.content) or template.content.cloneNode()

Can UMAs track avg # usages too, not just number of pages that use a feature?

That's nice idea. I think we can achieve that with custom code.