I don't believe the claim was that it adds no complexity, but was saying it adds no complexity due to the reasons mentioned. He then goes on to write a paragraph about why that is the case, which you could of course respond to.
To add more, we are talking about a hybrid scheme that uses both a post-quantum algorithm and a classical algorithm, where the former is much slower than the later and a mere hybrid doesn't make much difference in performance. And this scheme will be presented as a single choice in cryptographic protocols, so if the protocol already supports multiple cryptosystems it should be just fine to adopt.
DJB also explains needs for hybrids as follows: we don't yet have any post-quantum algorithm that we can confidently use as a direct substitute for well-established classical algorithms. So for now we need to consider possibilities for QC becoming real and PQ algorithms broken, hence hybrids.
