Which engine would need to learn this? The web browser (Fennec, MicroB, or whatever) receives coordinates where the input device clicked after which it decides to act upon this input, or not. So it already has to do that. If the hyperlink top left is at X = 50, Y = 30 and you want to zoom in if its 20 pixels off then if the touched coordinates are 50 - 20 >= 30, 30 - 20 >= 10 it should zoom in. That counts for every corner, but it also already has to do this in order to know a link is clicked. So it doesn't seem to be that much more complex than currently except that now there is also zooming which indeed does cost some resources, but with a hardware 2D/3D driver and because furthermore no other applications is focussed upon, this is neglectable.