Route optimisation isn't exactly the same as travelling salesman, though, is it?

Travelling salesmen is about visiting ALL nodes. Are they claiming that's what the bees do? Visit every nectar source on every run? And travelling salesman doesn't have that much to do with network topology unless your packets have to visit every machine in the network on their way. I think you've confused exactly what that problem describes and it's relevance to the rest of graph theory.

And the whole thing sounds dubious to me. Route optimisation in biological systems is pretty damn simple. Ants are a much better example, with their scent-trails that gradually fade but are reinforced each time they are walked over. That instantly provides every "junction" with a strongest scent on the newest (and thus, presumably, quickest-to-traverse-while-carrying-sugar) route, which then gets reinforced again and again with every ant that takes it and any wandering either provides a better route (stronger scent) or a worse one (less scent) that is immediately taken account of. They virtually perform A* pathfinding using things like that, except a bit more random.

And path optimisation in most things is quite simple. Find ANY route that works. Now wander around. If at any point you get to a point nearer the goal taking less steps than the final route, that becomes the best route (or new segment of an improved route). It's pretty clear from things like that that you can't just "look at an unknown universe and say what's the best route" without literally evaluating every single path (even if you discard most of them for being stupid), which is why lots of pathfinding and travelling salesman etc. is so time-consuming to solve.

Bees aren't doing anything special here. Probably barely worthy of study. You mess their routes up and then they find ANY route to the nectar and once they have that kind of triangulation the rest is just "tightening the string" (Imagine a bee trailing a string connected to the hive. When it finds the nectar, it ties off the other end of the string to the nectar. Now all path-optimisation is is "tightening the string" around the obstacles in its way and, sometimes, re-rolling the string around some particularly bad obstacles so it goes by a "tighter" route. When you unroll the garden hose to get to the back of the garden fence, that's the closest thing to optimal-path-finding you will do. The travelling salesman problem is like saying you need the hose to go by a ridiculous route touching X number of obstacles spread at all corners of your garden and STILL get to the fence at the end)

Personally, I'd be interested if there was not some more optimal mathematical way to *literally* tighten a string using physical mathematics and an appropriate "ideal space" to solve these problems by solving equivalent physics problems instead of pure graph theory analysis. But travelling salesman is another matter entirely where you have no choice but to check every permutation of nodes and see which gives the least total with the optimal route between each node.