They probably do because the effect can be pretty closely recreated with a displacement map in SVG.

I agree. Actual simulation would be an inexcusable waste of resources when the end result is simply indistinguishable from a normal distortion filer. Specially when their newest flagship has a smaller battery.

If you assume a fixed depth “behind” the screen and a fixed eye position, a lot of the math shakes out. It’s overkill for button backgrounds, but the actual implementation of a simplified simulation isn’t as computationally heavy as I think you’re imaging.