Not quite. The 2⁹⁶ private keys in question will produce 2⁹⁶ different public keys. It is the public keys which undergo RIPEMD-160 to produce the same output which will encode to the same address.

So different private keys, different public keys, same address.

There is no reason that two private keys which give an address collision for their compressed public key would also give an address collision for their uncompressed public key, and vice versa.