Honestly, I've been trying to figure out its use, and my optimism has faded. Can someone explain its usefulness?

My confusion is this:
1) Let's imagine you have three nesting levels, and you're in the third—the second subnet. Judging by the name, you'd expect to see the view of the previous subnet, but no—clicking the button takes you to the very top of the nesting level. But usually, I need the view of the parent subnet.
2) It's not just the viewport that changes; the node tree itself flies upward. And I have to manually click back to where I was. Or press the "back" button.
3) Perhaps a bonus would be to see the global view and edit the object in the subnet. But the active node handles are not available in the viewport in this mode.
Ultimately, it seems to me that simply assigning QuickView hotkeys to the nesting levels you need looks much faster, more convenient, and more functional.

it likely just follows parent context of the current one (not just a single level up within the same context)
so if you are in /obj/geo1/... no matter how many subnets deep you are still in the same SOP context of the geo1 node, so the parent context is /obj

this is likely a very old feature and also it's a very generic property of panes of any type, so there may be some various uses for various workflows, not sure how useful it is for viewport panes

the fact that the network editor also jumps when this is toggled on scene viewer pane (when they are not linked) seems unintentional to me and I'd also attribute it to this being an old feature

I think the practical use is probably limited for the way most people work in deeply nested networks. Personally, I’d only keep it in mind for cases where I actually want to step out to the broader object context. Otherwise, a custom hotkey setup sounds much easier for day to day navigation.

Follow Parent Context is built on a very old feature in Houdini, one that predated the existence of subnets, I'm told. The Texture Material Library motivated fixing/updating it, but the intent is you can work in a sub-context such as Copernicus, while keeping the Viewport viewing the parent LOP network; and it lets you switch the viewport back and forth easily. And regardless of how many subnets deep you might be working, you can still view the parent context (such as the render in LOPs) while still working in COPs.

Pinning isn't a complete solution because as soon as you un-pin, you have to navigate back and re-pin the viewer. So Follow Parent Context helps you bounce back and forth, as viewer states are an important reason to be able to bounce back/forth. This is demonstrated here in the H22 launch video: https://youtu.be/Arzhko0h8xk?t=6245 [youtu.be]

Hope this helps, we'll clarify the docs a bit too.