That’s a good argument, and as a fan of permacomputing and reducing e-waste, I must admit I’m fairly swayed by it.
However, are you sure JPEG XL decode/encode is more computationally heavy than JPEG to where it would struggle on older hardware? This measurement seems to show that it’s quite comparable to standard JPEG, unless I’m misunderstanding something (and I very well might be).
That wouldn’t help the people stuck on an outdated browser (older, unsupported phones?), but for those who can change their OS, like older PC’s, a modern Linux distro with an updated browser would still allow that old hardware to decode JPEG XL’s fairly well, I would hope.