This article is explicitly spun to promote ARCore as "Android scale" and leads with a sentence pointing out that Android is on two billion devices.
You're right, their ambitions are much lower, but it's fair to point out their actual goals aren't anywhere near the image they are trying to project and they really shouldn't be bragging about it being "Android scale" when it's nothing of the sort.
Most likely someone will just build and offer an abstraction layer on top of ARKit / ARCore and solve the problem that way.
https://github.com/google-ar/three.ar.js
That's assuming you're okay with JS, but if you're going cross-platform then it makes sense if you need to have interactive scenes.
The biggest performance bottle-neck is the JS itself, specifically stock three.js is quite inefficient. However, that's just a matter of optimising three.js, or using WebGL directly if you need more fine-grained control. You can get a lot done in a browser these days.
Maybe it is a consequence of using JS, I have never researched it.
Therefore it's fine that they get only 5% to start. I assume it's going to turn into 10%, then 20% quickly. For some perspective the iPhone only has about 15% market share globally.