The most important use for teaching Roman numerals that I see is learning that bad abstractions (here: a number system) can be made to work, but they still suck. The solution is to find better abstractions. Maybe it inspires curiosity in how (well) modern maths works.
> bad abstractions (here: a number system) can be made to work, but they still suck
The only really bad thing about Roman numerals is non-positionality, and it kinda follows naturally from them being merely transcriptions of the states of the 5+2-beads abacus that was popular back then. If only the norm back then were the 10-beads abacus... alas, the history is what it is.
