Nice to imagine, but the speed of light is too slow for a lot of low latency applications. After you allow the infrastructure some latency. A dedicated line might work, but without the network that is expensive.
Is it? Looking around, it seems average visual reaction times (VRTs) are on the order of 250 ms, and even the fastest VRTs are well over 100 ms. A study looking specifically at medical students for auditory and visual reaction times ("A comparative study of visual and auditory reaction times on the basis of gender and physical activity levels of medical first year students" [1]) also seems to support those numbers in the medical context, and while experienced surgeons could be expected to be somewhat faster that paper also links to research indicating hard biological minimums:
>"Researches by Kemp show that an auditory stimulus takes only 8–10 ms to reach the brain, but on the other hand, a visual stimulus takes 20-40 ms."
If we want to be quite conservative and aim to keep latency to single digit ms, so sub-10ms, then the speed of light in standard fiber gives us an RTT limit of around 1300 miles. Going more conservative and assuming actual route having to essentially follow the legs of a right triangle, that still gives a radius of around 460 miles, with a set of reasonably conservative assumptions. That seems plenty good enough to cover an enormous amount of work-from-home (or at least work-from-different-location) when it comes to actual existing medical practices, where I doubt most doctors at hospitals live >460 miles away.
I would agree that there are real issues with the idea, but more in terms of the risks events like network disruptions than the speed of light.
----
You argued that when we compare the average VRT additional latency is not large. But you should have argued that there was enough buffer between average VRTs and the time needed for a surgeon to react to allow adding to average VRT.
i.e. If we imagine driving a car with a video camera, we don't really care about latency vs reaction time. We care whether latency + reaction < accident threshold.
If you look at the paper I linked, it includes not just the mean but the standard deviation, which is on the order of 10-20ms. That means we are already by definition accepting that kind of variability regardless does it not? If +/- 10ms was critical, it'd imply we should be filtering for that already but it doesn't appear that's the case at all. Furthermore, hunting around for other research on the subject indicates far higher variability than that is also introduced by standard stress factors (lack of sleep, overload, distractions, etc). All of which are strongly present in existing medical practice as well. To the extent working remotely might reduce some of those it could in principle even effectively cancel out a 10ms penalty.
I stand by saying it's not at all clear that sub-10ms (or potentially even higher) would at all be the critical factor preventing remote work. Speed of light seems to be far less important then factors like total network disruption, robotics reliability, etc.
Mecheng tasks could be async / queue-based and latency wouldn't matter too much.
Regardless, I think it's really important to keep your bias for action high during this period of time.