The bumpy road to reference-count protection in the kernel

That seems an unhelpfully strict requirement. Today, not the whole developer base is comfortable with the kernel being written in C - some would be much happier with a language that had features to make it easier to write correct code and harder to write buggy code. Of course there are also people who would be much less happy with anything other than C. Any choice of action or inaction will attract some people and alienate others; it'd be best to consider whether, on balance (accounting for the effects on current developers and potential future developers and the objective benefits of the language itself), that would improve the long-term prospects of the kernel.

But I expect the people with the power to make those decisions are nearly all people who'd be less happy with any other language, and (since they're human) they're obviously not going to choose to alienate themselves for some hard-to-quantify long-term benefits to the project, so in practice nothing is going to change.