They'd stop working earlier on bad days instead, so the leisure time total would not decrease (well, assuming roughly equal numbers of good and bad days, with median days being the most common.)

This assumes that the effect of leisure time on well-being is merely a function of the amount of leisure time you have. That seems unwarranted: you can't simply take a long break and then spend the next year working with no leisure time at all, even if that gave you an equal amount of leisure time as a scenario where you distributed it more evenly. Likewise, it seems intuitively plausible to me that if you

a) take the day off early as a result of having made a lot of money

b) take the day off late after you've worked through a bad day and gotten a feeling of deserving that time off instead of quitting early and not accomplishing anything

your subjectively experienced leisure time quality ends up being higher than if you'd distribute the leisure time otherwise. (At least for some people - admittedly, I think I'd myself feel better if I did things the way you propose.)