> In practice, we don't know if miners would actually include cheap transactions if it makes their blocks significantly bigger. Their incentives seem to be against it, since bigger blocks have higher orphan rates.
This is incorrect in fact. The Google search term you'll need is selfish mining. It's a well studied phenomenon that for sufficiently large miner they stand to benefit from having larger blocks.
I think your last paragraph is right.
Right now there's a default transaction inclusion policy in Bitcoin Core that only includes a small number of low to zero fee transactions. Under current conditions zero fee transactions have to wait a while to be confirmed.
There's no proposal to change the default allowance for below standard fee transactions if the block size increases, so assuming cheap means zero or very low fee the increase in size will come from standard fee transactions which means increasing capacity won't drive down existing fees.
But if blocks are not consistently full and there's no significant backlog, there is room for cheap transactions. This has been the case up to now. If the block size limit were to be increased, the assumption goes, there would again be room for cheap transactions - until they fill up again.
In practice, we don't know if miners would actually include cheap transactions if it makes their blocks significantly bigger. Their incentives seem to be against it, since bigger blocks have higher orphan rates.