@StC is using the Socratic method on us here. 
Your answer highlights one of the problems of the current network=**n scheme, which is that it assumes a specific use case – one that has been important for OSM historically – but masquerades as a more general-purpose scheme. highway=* also exhibits this problem but to a lesser degree. (Or to a larger degree, if you count by the number of forum threads about it in the last few months…) Ironically, multiple cartographers have pointed out that the current usage of network=**n may not even be as well-suited for this single prioritized use case as we have thought.
Indeed, QRank would be an interesting exploration. It’s based on page views across Wikimedia projects. The best part is that the original question that started this thread goes away: we don’t have to tag anything, and each data consumer can decide the thresholds for themselves.
Normally, when QRank comes up in the context of OSM, cartographers immediately assume it’s the holy grail of cartography: an automated mechanism for ranking populated places by importance. If one does it too naïvely, as in this experiment, the result is a simultaneously hilarious and depressing conversation starter. Most maps at zoom level 4 shouldn’t ambush readers with decommissioned fighter jets, crash sites, and massacres, as this one does. After all, QRank prioritizes all the interesting features, but the set of important places may actually be quite boring by comparison.
On the other hand, if you use QRank to rank a more specific class of features, such as hiking routes, the results may be more predictable, especially in combination with other heuristics like distance and annual traffic. In principle, one could also do the same with entire networks of routes, if that’s the kind of map they want to make.