You really do. Which features of Roman numerals are better than Arabic? Is it best for computers to use binary, and does mathematics really support such reasons? Of course you can reason with mathematics to support the validity of your reasons, but the math is likely second-hand reasoning. I feel like everyone likes to pick up all the arbitrary and normative reasons handed to their ass and perceives them as inherent features of mathematics. You like to claim that's not true... whatever. This also occurs in... everything else. It has a lot of stupid side-effects, but it's just how humans build things. Human nature, though awesome, can often be annoying.
Setting aside the vaguely mathematical angst about human nature, I know a couple of things about computer implementations of number systems, and I can tell you are confusing a number of different levels:
- Formally right or wrong. All number systems "work" in the sense that they can represent numbers and be used for computation, but some are better than others.
- Special good and bad properties. Properties are formal, but what makes them a feature or a problem are technological factors. For example, number systems with only 2 symbols are a good match for electronic circuits, but not for many types of mechanical calculator. Inventing different representations using 2 symbols is an interesting mathematical challenge driven from concrete applications; nobody thinks "inherent features of mathematics" are involved.
- Fitness for a purpose. Nobody in their right mind thinks there is a "best" number system; it depends on what has to be computed and on a number of budgets and constraints; all applications require trading off representation size and/or electronic circuit size against computation speed (e.g. avoiding carry propagation in addition by storing 2 bits per digit) or initial and final conversion effort against computation effort (e.g. enduring BCD arithmetic because for few calculations it might cost less than converting to and from binary). Normal people don't have any particular "arbitrary and normative reasons" affecting these engineering choices; using bad technology (e.g. most arithmetic with Roman numerals) is usually the result of honest ignorance, not of epistemological biases.