I think a certain amount of "horses for courses" caution is called for here. Some skills seem much more suited to learning by doing than others. Even within a particular field of endeavor, some skills seem more suited to learning by doing than others. Didn't various militaries converge on the practice of sending people to school to become artillerists and celestial navigators even as they continued to honor learning by doing in other specialties? If you want to rise in politics or design user interfaces, it may work pretty well to learn by doing, while studying only as the need arises: as far as I know, there isn't that much formal knowledge that's foundational for that kind of work. There's still lots of valuable stuff to read and learn from, but it doesn't seem foundational: trying to read a UI book without studying the prerequisites doesn't seem to be nearly as confusing as e.g. trying to read a physics or engineering book without studying the prerequisites. And math-y stuff beyond the usual public education level seems foundational in a fair number of specialties, not just physics and engineering. If you want to price portfolios of derivative securities or write software that recognizes faces, a couple of years spent on math and/or on related fundamentals of software (algorithms and data structures, various other programming skills) may be a good way to start.

I'm not suggesting those man-years are necessarily best spent at school. I notice that most of the valuable formal foundational knowledge tends to be open to self-study if you're motivated enough, and I have run into people who'd learned a lot of them for themselves before they got out of high school. (E.g., my freshman roommate at college.:-) But I am suggesting that in school or not, some of those man-years are best spent before you need to rely on the knowledge: learn a lot of what you need to know about spherical trigonometry and log tables or gunpowder mixes and metal fatigue before you need it, not as you need it.