Very interesting observation. Here is just my thought.
If these orgs are forming a consortium and a fabric network at the very beginning, I suppose they somehow agree on some philosophy about consortium network and applications, that certain data will be limited to a number of orgs due to business nature. This by no means take out some "rights" from an org as all orgs will have this right, i.e., having private data with some selected orgs.
While in v1.4 chaincode instantiation (where private data collection is defined) can be done by a single org, v2.0 claims to be more democratic, in a sense that each org can approve or not a chaincode definition. In such a case, an org will know that it doesn't belong to a private data collection when it makes the approval.
Yes, in this arrangement, the private data collection definition is known to org3. "Org3 knows that there will be something unknown to org3". If org3 is not happy with that, I would say org3 shouldn't continue to be a member of the consortium network.
Or in real life, there will be many collections existing: org1-2, org1-2-3, etc. We don't need to remove org3 from org1-2-3. Rather, simply use org1-2 is good enough. But again, org3 also see the definition of org1-2. :-)
Once again, thanks and it helps me thinking more.