I am not sure the benefits of a protocol aligned with the needs drastically outweigh the lack of network effects.

Every collaboration-centric software start up designs their own protocols in a similar way you described. Most of them fail, despite the massive capital investments they usually receive.

I think a better idea is to reuse a less-optimized protocol but with large network effects (and hopefully help build those network effects) to avoid being replaced by the next startup with millions in cash laying around.

It's like JavaScript. One of the worst languages ever invented. But one of the most used ones, nevertheless.