I agree. I think for this concept to work the party that wants a fork will have to provide the
fork within their own domain name identity and at their own server cost. The domain name
identity is needed so that better or worse forks can acquire their own reputation independent
of the parent node. As far as server cost is concerned, those wanting a fork should either
have to put their money where their mouth is or find a sponsor (e.g. on the sourceforge model)
willing to host them. I think it somewhat unlikely that a hoster of a useful data collection
with a free software license wanting more flexible use of this data would want to write a
blank cheque in connection with the cost of supporting community experiments on code
functionality.
Using virtual machines or alternative hosts, for forks with their own distinctive identity,
where there is some accountability for server costs and unacceptable software features is the
way to go.