I agree that the ARPA model is not suitable for all (or even the majority of research) and the act doesn't inspire confidence that shifting the NSF towards a DARPA-like model would do well.
I have two thoughts that might erode your thesis a bit, that 1) I think you're overestimating the amount of time that faculty spend on writing grants. We like to complain about it, but I've tracked my time to the minute over the past 7 years, and grant writing (both the proposals and the reports) is about 2% of my work time. 2) You might be misunderstanding indirect costs like many people, where you say "Universities can take more than half of grant money as administrative overhead". Indirect cost math is funny, but in order for administrative overhead to be more than half of the grant, the indirect cost rate would have to be over 100%. I can explain more if you're interested.
Obviously we're both biased by our job, but it's still useful to read your perspective.
I would characterize my position as less “pro DARPA and anti NSF” and more “I think on the margin the world needs more DARPA-like activity more than it needs more NSF-like activity”