Markets don't discover "the right" (i.e. social welfare maximizing) resource allocation, they discover, under ideal conditions, a pareto-optimal allocation. This is a very weak condition: "Jeff Bezos owns everything" is pareto-optimal. You need redistribution if you want to turn that into a socially optimal one.
In realistic conditions, they don't even do that: market mechanisms will reliably underinvest in things with positive externalities, and the larger the externality the greater the degree of misallocation. The positive externalities of research are enormous compared to the value captured by the researchers.
> The positive externalities of research
... are only relevant if it's actually research, and not something that looks a lot like research without actually being so.
If anything research is clearly over-funded right now, especially in some fields like medical research. That's why so many scientists turn to fraud in desperation. There just aren't enough genuine discoveries to go round.
> encouraging replication, self directed research and significantly more 'pure' 'blue sky' research
Replication is the opposite of self directed blue sky research, so how do you suggest it be encouraged whilst simultaneously reducing the power of the money givers even further? BTW, many bogus papers will replicate whilst still being wrong. Replicability is not a synonym for good science.
Really, this is a very common claim that needs to go away. The solution to fraud and incompetence is not to give incompetent fraudsters even more freedom and money.