This kind of decision is already routinely made by committees awarding NSF postdocs, Clay fellowships, etc. What's wrong with the process they use (and independent expert panel)? Do you think they select the wrong mathematicians?
cperciva's criticism of this process is, as I understand it, that the grants they give are too short, and that this, together with the selection criteria, creates a too-strong incentive to pursue short-term, low-risk results.