The speed of light is (currently considered to be) the fundamental limit.
Let’s try some figures:
The Earth’s circumference = 4 x 10^7 m
Speed of light in vacuum, c = 3 x 10^8 m/s
Therefore, the minimum time for a light ray to travel between two points on opposite sides of the Earth (say, by a perfect optic cable of minimum length),
t = 0.5 x (4 x10^7) / (3 x 10^8) s
= 67 ms
But in practice, the path won’t be a great circle (there will be doglegs) and optic fibre transmits at something less than c. Not to mention that there is various relay equipment along the path.
So I’d be surprised if that figure can ever be reduced below 100 ms.
So the Verizon latency figures, say 150ms for US-Australia, is not all that much above the fundamental limits.
And recall that the graph I posted is indicating a maximum of about 50ms acceptable latency for FPS games (because those enemies that you’re trying to shoot move fast).
The inevitable conclusion is that for some SAFE applications it is going to be necessary* to use out-of-SAFE channels for some of the data, with communications within SAFE for the setup. Telephony will be the big one.
* Unless: The devs (core or app) might “optimize” by introducing a ping test to allow an app to, in effect, sort prospective partners according to geographic proximity. But such discrimination would have to include all nodes along the SAFEnet path and not just the end nodes. In which case SAFE will no longer be geographically agnostic. Is anyone actually working on that?