Thanks for the sharper video. I may still misunderstand what you are concerned about, but I believe it is about two separate issues:
(a) The position of the gate cursors is obviously limited to the time window shown on the screen. So when you zoom far in (100 µs/div and faster in your example), the cursor positions get "clipped" to the screen boundaries. This makes sense to me -- after all, you set the cursors because you want to measure in specific sections of the signal, based on your visual observation of the signal. So why should they be set somewhere out of sight?
(b) When you zoom far out (50 ms/div and slower), the "green" measurement values, i.e. the inter-channel delays, are no longer correct. At 50 ms/div they are not measured at all; at 100 and 200 ms/div they come back, but are incorrect. I find that unexpected too, since the scope supposedly makes all measurements in its full-resolution data buffer, so they should not change. (And the CH2 width measurements in fact don't change.)
In summary, I see no issue with (a), and an unexpected behaviour -- although not highly relevant in practice -- with (b). And I would love if you could describe such issues similar to what I tried for (b), rather than letting me stare at a video and take my best guess...