Furthermore, much of NASA's success is due to the mix of scientists from different fields, universities, and business working on one problem. When the best of the best come together, it results in a cross polinatiom of ideas. It's not just being in the same building or sharing a cafeteria while working on disparate grants, but a day to day sharing of knowledge to tackle a single problem.
A Caltech scientist who had worked with DOE and NASA labs made an offhand comment about the kind of work that those labs are suited to do that stuck with me - he said they should be trying to solve problems that are "national lab hard" - kind of a play on NP-hard.
What he meant is that these environments are suited to solving complex cross-discipline engineering/science problems ("make the right measurements to understand the climate system", or "find the processes and materials to make air travel safer", or "put a huge infrared telescope in space").
These are not problems that are suited to a university research structure. They don't get solved with $300K/year grants, small collections of discipline-focused PIs, and transient grad students.
Or if you prefer a programming analogy, what if all software teams only built what they were explicitly tasked for and could not collaborate on the tools used. Managers _might_ get their deliverables faster but I suspect by not developing tools incidentally we leave out future productivity gains.