Suggesting that an application should stay within a 128KB limit is akin to saying I enjoy playing games in polygon mode. Battlezone was impressive in the 90s, but today, it wouldn't meet user expectations.
In my opinion, initial load time is a better measure of performance. It combines both the initial application size and the time to interactivity.
Achieving this is much more complex. There are many strategies to reduce initial load size and improve time to interactivity, such as lazy loading, using a second browser process to run code, or minimizing requests altogether. However, due to this complexity, it's also much easier to make mistakes.
Another reason this is often not done well is that it requires cross-team collaboration and cross-domain knowledge. It necessitates both frontend and backend adjustments, as well as optimisation at the request and response levels. And it is often a non-functional requirement like accessibility that is hard to track for a lot of teams.