How many mathematicians can you let loose on long-standing physical problems (qualitative dynamics of the large-N body problem, the freaking turbulence motion of fluids, etc.) -- at some level of "big bet" that frees them from staccato publication pressure -- with the money spent trying to find gluinos or some such ill-developed theoretical construct?
It's hard to imagine the output of that wouldn't be amazing.
After that, how do you separate the promising mathematicians from the lazy and the crackpots?
This is the same problem that a "Manhattan Project" to cure cancer or what have you always runs into: It's easy to see where to get value from the first dollar, but the 30 billionth dollar likely costs more than a dollar just to figure out how to productively spend it!
"Fortunately", experimental particle physics doesn't have this problem, since you can always use that next dollar to build a bigger collider.
I say, start a program with the Erdös-2 people and as it develops let these hire Erdös-3 folks.
so we should probably allow the smaller Erdös numbers to be inherited through primogeniture, to make sure we still have an identifiable class of good mathematicians to give money to.