In a situation that affects so many people, it may seem unwise to do experiments and try things where we can not say with certainty what effects they might have, though we admit they may be substantial. The ethics are very dim on that front. But the alternative has to be considered. We do not have 'generally accepted treatment' to turn to and rely on. We only have other experiments, albeit while not calling them experiments and not treating them as if their outcomes are in doubt, simply forging ahead with blinders on. I can't imagine an argument that would justify that as more ethical than earnestly doing experiments.
This is analogous to why econometrics is bullshit: lack of stationarity in the real world.
This is amplified by nonlinearities in the system. So even if things today were broadly comparable to, say, the 1990s (and they're emphatically not), what if there was this little thing somewhere (like an obscure accounting regulation) that led to massive differences in the outcome of the experiment?
Tbc, I think there are better and worse ways to deal with the problem, just because we know so little does not mean everything is equally meaningless.