In terms of scalability, I don't think it is possible to increase the number of transactions in a block by one without increasing the total block size by some constant. In other words, the total time incremented block size will be at best O(1) in relation to the number of users who have at least one on-chain transaction during that time increment.
That's correct. A decentralized blockchain by design,
'scales' extremely badly.
Because instead of just requiring to add a few more servers (traditional 'scaling' method in computing),
every single node / 'server' on the network would need to drastically improve their hardware.
Meanwhile off-chain solutions allow us to handle a much higher maximum number of transactions per unit time per processor. Whether that will be LN or not, remains to be seen. It's possible that there will be a whole ecosystem of off-chain protocols that are compatible with each other.