Is this still (or has it ever been) actually the case?
There is "automatic unrooting on destruction" https://developer.mozilla.org/en-US/docs/Mozilla/Projects/SpiderMonkey/JSAPI_reference/JS::PersistentRooted and since PersistendRooted variables are rooted in the JSContext, they should be deleted when the page is closed, i.e. as intended, no? Perhaps this was not the case in 2007 but now is?
(Does this relate to the MEGA TODO from rP141 in Destroy of IGUIObject.h?)

^

Seems like they are associated with the runtime as opposed to the JSContext.

These roots can be used in heap-allocated data structures, so they are not associated with any particular JSContext or stack. They are registered with the JSRuntime itself, without locking, so they require a full JSContext to be initialized, not one of its more restricted superclasses.

Maybe I am misunderstanding something here?

Ah yes, I was already equating JSRuntime with JSContext https://developer.mozilla.org/en-US/docs/Mozilla/Projects/SpiderMonkey/Releases/45#Future_Direction

The patch to change it to JS::Heap is simple, but measuring the memory leak isn't.
valgrind + massif is something that was slow and I suspected it won't show conclusive evidence easily.
I have shown allocated memory using ps and called this function GetGuiObjectByName on hundreds of thousands of GUIObjects, but the allocated memory didn't seem to increase.