> If they didn't focus on toy math problems they would have to do other stuff like endless, increasingly complex form validation, filesystem manipulation, or things like graphics that involve external libraries-- it would just become hard to manage for code-golf scenarios, I think.

It would also make it hard to use the exercises in some languages. For example, awk or retrocomputing languages like JOSS or BLISS.

There's definitely a need to have good little or mid-sized projects which do call for library-wrangling, though. They help get programmers acquainted with the culture surrounding the language, and what kinds of problems other people using the language are solving.