Note that not having an RI does not mean a spec can be released without a ready to go implementation. I believe the spec process needs to include working code, but that can be from multiple implementors.
For example, with the development of HTTP/2 at the IETF we had several implementation developed during the process and interoperabilty was continuously tested. In fact the protocol and process included support for draft versions which we tested against. Ditto for websocket, which went to draft 13 implementations before final.
Just having an RI means that you have to try to generalize from a sample of 1, to say the spec is good. Yet any RI is going to have its own history and focus that may make it's behavior not generally applicable or that it is not able to detect some issues with a spec.
So having working implementations during spec development is vital, but once the spec goes final there should be no implementation that is any more special than any other. Each implementation should be evaluated on its own merits against the TCK and specification document, not against some other vendors implementation.
Cheers