Sometimes a reference implementation is intended to be just that, and nothing more—a reference, for other implementors of other systems to crib ideas from, especially as pertains to how the smaller ideas cohere architecturally in practice (which is hard to communicate by just writing papers about the individual ideas, or building them as standalone libraries, or building them into other systems where they must interoperate with components that don’t cohere into the desired architecture.)

Or, to put that another way: sometimes the best textbook is the codebase of a good implementation. And some people are more interested in writing those textbooks, than in personally getting their ideas into production systems. Especially because it’s far easier to refine the ideas in the context of such a coherent presentation of them.

Someone else, who reads such a “textbook” codebase, can put the ideas to work. But the ideas themselves will be better-developed for having had time to stew in the same pot as other novel ideas. This serves the people who put the ideas into practice as well.