One reason your proposal isn't happening today might be because we don't want tons of doctors choosing to enroll in these "field" randomized trials, if they might be experimentally treating people with drugs that aren't safe. We want to know about safety first before unleashing on the world, even in a randomized trial context. And since the FDA bundles verification of safety and efficacy together, so we can't do your proposal without unbundling them.
Now, imagine that after safety is verified to a 1 or 2-star level, we don't make the drug approved for general use, but doctors can freely enroll in randomized trials as you proposed? That way we're getting the efficacy data quickly but safely.
Maybe a minor point, but we do have to unbundle safety from efficacy in some sense to get off the ground here. The question is, what exactly is allowed to happen after safety is established but efficacy is not yet established?