I've often thought this approach makes sense and wondered why it didn't seem more common.

The two challenges are gathering the information and interpreting it. Logging which links are clicked is obviously easy but not necessarily a good indicator of interest (for instance I'll often Ctrl-Click a number of links then read them rather than click one at a time until I find what I want).

It sounds like they are using a representative sample of users with a plug in to allow them to collect more advanced information (such as time spent) which is an interesting approach.

I can imagine that there may be false positives though - for instance a site that looks like it should contain the desired information (but actually doesn't) might occupy users for longer than a page with the answer they want right at the top.