Any time you're randomizing, it is going to be an inefficient use of public funds. I do think that some sense of merit, at least as defined by the academic discipline, can be based on a proposal and what someone has published in the past.
The problem is that if an academic discipline is just getting it wrong in deep ways, then there are benefits to stepping outside that. An alternative to that, if I could suggest one, might be having another stream that's easier to apply for but offers relatively low amounts of money for low-cost research. SSHRC has tried to do this, I think, with the lower stream of insight grants. Keep in mind that I'm thinking about social sciences and humanities, where some research has pretty low costs.