Overall I think this is an interesting idea and one which could be used to address the apparent dichotomy between crunching popular projects with an obvious benefit to humanity and those which are more lucrative through being less popular, but are also less beneficial. There are a couple of points which I think would need to be addressed however.

First, as @jefpatat pointed out, using a simple grouping system could easily lead to unfairly biased rewards for a small group of, or even one, project. Personally I would therefore favour some sort of per-project weighting system instead.

There is also the problem that the benefits of any given piece of research are not always obvious until after the fact, sometimes decades and occasionally centuries later. This is a potential pitfall for any weighting method, and whilst it isn't an argument against doing it at all, it is an argument against penalising projects too severely purely on the basis that their societal benefits are not currently apparent.

That said, this is absolutely something that would benefit further consideration!