I definitely envision tech allowing transactions to be merged/skipped and whatnot between large routing nodes that are processing lots of transactions in both directions through a channel

483 pending payments at once is a lot, even considering they might take ~500ms each, that's still 1000TPS per node. the theoretical limit is likely smaller though, because of latency involved in filling and emptying all 483 slots.

As we scale lightning, things like channel factories will mean that there are still way more channels than the TPS we need. There are different ways to model this, and it's yet to be seen how different channel constructs will become more and more used or will die off; but just ballpark ~20TPS per channel is very flexible considering most people do <0.001 TPS in their daily lives.

In a model like 6 degrees of separation, we would only need to route through ~6 nodes to get to any destination, something that will only take a few seconds to do. With things like channel factories, this number will likely be closer to 4 or 5.

I believe you greatly overestimate how many transactions need to be processed through one channel at a time to achieve worldwide scalability. We will have a lot of channels by that point.