What exactly do you think will be tested in that artificial environment when the environment itself (with what I'd hazard to guess quite a list of specifics, many of the "subtle and implicit, but influential" kind) is a central variable of the test?

If an observational study of this kind is deemed hard/improbable/impossible to set up - how hard do you think it is to get the artificial setup right? To account for - or even identify - all variables and preconditions? This is so easy to mess up and end up with basically meaningless data. (Yeah, I know - that doesn't mean it won't be published. Everybody raise your hands and shout "Replication Crisis"! :) )