The question is, if on-chain-scaling is only a mediocre improvement at the best given current capabilities, then why are we pursuing it at all? a 100-500% increase probably won't make much of a difference since it still won't be in the ball-park needed to be used for small & instant transactions anyways.

I'm not saying that improving the on-chain scaling protocol isn't a good idea, only that unless someone introduces a radical improvement (nothing of the sort that has been suggested up until now), the focus should lie on developing other aspects of the protocol (instead of wasting time & resources on mediocrity).

Make improvement on things that can yield significant improvements, not mediocre improvements. Value prioritization.