That's more an experimental vs theoretical boundary. It's like hardware vs software, the former just has extra physical constraints that introduce significant o

That's more an experimental vs theoretical boundary. It's like hardware vs software, the former just has extra physical constraints that introduce significant overhead that is most efficiently managed in a hierarchy so there are very few practicing PIs. AFAIK there aren't many (or any) renegade theoretical physicists/mathematicians working on teams at the LHC or ITER or similar sized projects, as an extreme example. Just getting plugged into the ecosystem and getting time to run an experiment is close to a full time job, let alone actually designing a meaningful experiment. Hell, even acquiring lab equipment for small scale experiments is a huge time sink, especially for the nicer stuff that's so expensive it requires labs pooling together and fighting over the allotments. I'm not so sure that PIs not doing any real work is the real problem though based on my reading of The Structure of Scientific Revolutions and personal experience. Science is about breaking new ground and I'm not convinced that seniority instills ability like it does with mathematics and engineering, where pattern matching builds on itself and its utility increases exponentially. Seniority in science does improve one's ability to navigate the scientific bureaucracy, but the problem is that the PIs accumulate power and the entire system calcifies around them. It was already such a noticeable phenomenon that Plank coined his principle in 1950 ("Science progresses one funeral at a time") when there were far more individual PIs who did their own work. Now we've taken it to the extreme where each subfield has a small group of larger labs creating an incestuous circle of peer reviewers, PIs, and postdocs who end up dominating the subfield just out of sheer numbers and prestige.