Earlier quoted context omitted.
> and hope they know someone The system has routing, and it turns out that it doesn't take much for the probability of a graph to be connected with low average diameter. See: The six degrees of kevin bacon. If lightning doesn't work for a particular payment, you can simply make a payment without using it, potentially by splicing out funds from one of your channels. Yes, Lightning has trade-offs. You have to be online…
> though if you don't like them you're free to not use it. No true: due to limited block space on the base-layer. I am not convinced you get any efficiency increase with the LN: just more difficult capacity planning because everything is suddenly so hard to measure. Sending a payment, whether on the first or second layer, will take a certain amount of: bandwidth, processing and storage. Even if fewer nodes are involv…
Today sending a transaction communicates 10 messages for each of ~100k nodes in the network. Once its confirmed, that transaction will additionally be sent once to every new host to join the network, forever.
Lightning sends a couple of messages back and forth among the nodes directly involved in the transaction... maybe 4. (The average shortest path length is about 2.8 currently).
So the marginal communication cost for a transaction is literally hundreds of thousands times lower-- even ignoring the cost to future nodes joining the network-- and this advantage grows as the number of nodes increases.