This experiment was real in the earlier era of the Internet. And the answer is: When censorship isn't universal, and there are lots of more or less free spaces to speak in, extreme views don't get concentrated and exist everywhere but at a much lower level of intensity. The mainstreamers are less ignorant too because they actually have some exposure to arguments and counterarguments against these views as opposed to just vilifying and banning the others.

Lots of other benefits to this model - names that people don't get trapped in extremist bubbles as much when they're not literally banned from everywhere else.