I do not disagree that on chain capacity needs to be more but just from what I can gather from my own observations that everybody seems to be building on L2, as long as,,, on L1 when they finalize the channels they can be sure of security and permanence. Which is still the biggest strength of Bitcoin as opposed to Ethereum and others all popularly using L2.
L2 and L1 are fundamentally different. L2 is suitable for smaller transactions while L1 is suitable for larger transactions, which makes far more sense. L2 cannot scale without the base layer being expanded as well. We will always have on-chain transactions when opening or closing the channel, which will happen far more frequently than you think. Lack of transactions on-chain will probably not be due to the popularity of L2 for years to come. There are still various hurdles to overcome.
I do think the level of security is high enough that it does not need to be much higher, just as you say to make it too expensive to attack. But it is not just about hashpower also but the diversity of hashpower that is important.
Mining is as decentralized as it gets, anything that favours economy of scale will also result in centralization. That is a fact. The security would primarily be regarding the costs of attack rather than how decentralized it actually is. If it is not desirable to attack at all, then it simply won't happen.
I agree that L1 confirmation (as I said to finalize the channels) will be happening a lot more often than people imagine,,, which is actually proven already with Ethereum (when looking at Polygon for example), when people have L2, it seems they are more willing to conduct more transactions. For example separating 10 items on individual orders as opposed to batching 10 items on one order, so the myth of L2 pulling away business from L1 is already not true (people still make the final order regardless).
What I meant about reduction of hashpower is that if people leave,,, generally only the smaller ones leave, those whose profit margins are not as good as bigger companies. So a problem might arise if smaller players keep leaving, and only big ones are left. So we can deal with hashpower reduction but diversity might be an issue.