The question to my mind is whether the value of attempting to make such estimates is sufficiently great so that time spent on them is more cost-effective than just trying to do something directly.

Can you give recent EA related examples of exercises in making quantitative estimates that you've found useful?

To be clear, I don't necessarily disagree with you (it depends on the details of your views on this point). I agree that laying out a list of pros and cons without quantifying things suffers from vagueness of the type you describe. But I strain to think of success stories.