Say the situation where a miner includes n transactions in a block or n+1 is equivalent in cost*.
This situation does not exist.
The cost is very real and not-insubstantial - the increased orphan risk which you completely glossed over above.
The marginal cost can - and should - be brought very low but it will never reach zero.
The amount of transactions a miner can process in a ten minute interval is finite, therefore it is scarce, therefore the service of including transactions in a block will always have a price.
QFT!
The marginal cost decreases with the block reward proportionate to the mining fee.
So while the n vs n+1 will never be equivalent, they should be expected to increasingly approach the risk/reward for orphaning.
This should then also be expected to increase the desirability of including the n+1 transaction in the block.
At some future time the transaction fees may be more than the block rewards, as block sizes increase and the block rewards decrease.
This future time will likely be
after a much longer term bitcoin price stability than anything in its history. This may well be for younger people than I to observe.
So far, bitcoin valuation has increased so much faster than block rewards decrease, and so transaction fees have decreased faster than block rewards. We should not expect this to be true forever, just for a long time to come, and also for block sizes to increment over time.
tl;dr The protocol works very well and these considerations are baked in. Bitcoin is quite secure in this regard.