Not a front end developer but I’ve always assumed it was ROI - nobody felt a smaller memory footprint would drive enough incremental revenue to offset the costs, compared to the other things they could be doing.

What seems to have changed is the size of the costs - LLMs are really good at this - such that you can now say “well, maybe the returns aren’t that big, in fact maybe they’re zero, but this is easy enough, so why not”

De minimus improvement has always suffered this problem - we’re all fairly sure it’s worth something. Maybe there’s even real perceptible value when you fix the whole basket. But making the case is hard.