I have seen an implementation that in signature verification time searches for OCSP responses signed by the CA validating the status of the CA itself, so my question is if somewhere (maybe RFC2560) says that you can revoke a CA and sign the OCSP response with the same revoked key?? does it make sense?
No where in the standards that I can recall (I contributed to 2560 and am an author to 5019) does it say you can not serve responses for yourself, that said as you point out doing so is useless since you cant trust the response in that scenario.
Are you sure that is what you are seeing though; remember that OCSP "good" doesnt mean issued it means "not revoked".
If you ask an out of scope question to a responder you may get back "good" even if the certificate was never issued. One theory as to why you are seeing what you describe is you are asking the ocsp responder for the issuing ca using a test tool even though no client should ever ask that question.
Also in theory in this case you would expect the issuing ca certificate itself to contain an aia:ocsp:url reference but to a different ocsp responder.
Anyhow, what you are seeing is not right as you point out I hope it is a problem with the test if not I would notify the operator of the responder to check its rfc compliance and interoperability as there are likely other issues too if this is the case.
I think that logic has a flaw related to who provides the OCSP service, namely the one and same CA. The OCSP service URI is provided as part of the certificate process and maintained by the CA. For your suggestion to work the CA would need to spend cash to support an OCSP service to state to the world it is not a valid CA.
Market forces suggest this will never happen. The same market forces that probably motivated the CA to get in the business and then take shortcuts which resulted in the need to revoke that CA.