In my opinion, the number of full nodes doesn't matter (as long as
it's enough to satisfy demand by other nodes).

Correct. Still, a high number of nodes has a few other benefits:

1) The more nodes there are, the cheaper it should be to run each one, given that the bandwidth and CPU for serving the chain will be spread over more people.

2) It makes Bitcoin seem bigger, more robust and more decentralised, because there are more people uniting to run it. So there's a psychological benefit.

Also, we don't have a good way to measure capacity vs demand at the moment. Whether we have enough capacity is rather a shot in the dark right now.
 
What matters is how hard it is to run one.

Which is why I'm interested to learn the reason behind the drop. Is it insufficient interest, or is running a node too painful?

For this purpose I'd like to exclude people running Bitcoin Core on laptops or non-dedicated desktops. I don't think full nodes will ever make sense for consumer wallets again, and I see the bleeding off of those people as natural and expected (as Satoshi did). But if someone feels it's too hard to run on a cheap server then that'd concern me.
 
My own network crawler (which feeds my DNS seeder) hasn't seen any
significant drop

It would be good to explain the difference, but I suspect your definition of "well reachable" excludes people running Core at home. From the diurnal cycle we see in Addy's graphs it's clear some nodes are being shut down when people go to bed. So if we have 6000 nodes on servers and 2000 at home, then I'd expect Addy's graphs and yours to slowly come into alignment as people give up using Core as a consumer wallet.