Easy. Split between the first 2 groups. Currently, too much money probably goes to the 3rd group, but there is non-negligible value to an established lab.
The problem is not who to give the money to, the problem is what research ideas to fund. Currently, funding is too risk adverse. Especially in the hard sciences, you have to be able to make claims that your research will have near-medium term economic benefit. This is a great way to allocate a portion of funding, but once a researcher has made a name for themselves in some way, they should be given to go-ahead to pursue more speculative ideas, and then not punish them immediately in the next round of funding if it didn't work out.