On Monday, October 5, 2015, Mike Hearn via bitcoin-dev <bitcoin-dev@lists.linuxfoundation.org> wrote:

As Greg explained to you repeatedly, a softfork won't cause a
non-upgraded full node to start accepting blocks that create more
subsidy than is valid.

It was an example. Adam Back's extension blocks proposal would, in fact, allow for a soft forking change that creates more subsidy than is valid (or does anything else) by hiding one block inside another.

Maybe I'm missing something, but wouldn't this turn into a hard fork the moment you try to spend an output created in one of these extension blocks? So sure, the block that contains the extension would be considered valid, but unupgraded validators will not update the UTXO set accordingly, meaning that those new TXOs can't be spent because, according to their rules, they don't exist.