I worry about the tendency of large institutions to coalesce power & apply that power toward their own biases. Since we’re trying to maximize variety, it makes more sense to lower the necessary resources to the point where many small institutions can perform their own versions.
Consider something like the Long Now Foundation — they’re a fairly strange group, with unusual aims, but for various reasons (mostly related to the personal connections of individual members) they have a huge amount of clout, and therefore they can’t really be as experimental as they need to be to achieve their aims. They have too much influence to be able to take small risks in public.
A better model is writing contests. For a little while, I followed a publication that did a monthly roundup of open writing contests, and I was surprised to find that each month there were hundreds, mostly funded by small universities or foundations, with prizes between $25 and $1000 and restrictions ranging from “entries must be below 100 words” to “authors must be native american lesbians from rural Wyoming over the age of 65”. Individual contests have very specific biases, but each one is small enough that the total bias across all contests is kept small.