Instead of "At the current overhead spending rates, has the research been effective enough?", I think the question should be "At the current indirect rates, has the research been supported enough?". My proposed reframing is because the indirect costs are meant to cover the costs of supporting the research.
A side node: If you're interested in gauging the effect of a particular NIH grant, the NIH RePORTER site is your friend: Any publication that is grant-supported will mention this in the acknowledgments. For example, our SGI UV300 (360 CPU cores, 10TB RAM, 20TB PCIe flash, and 4 NVIVIA P100s) was funded thanks to NIH grant S10 OD023452. If you search that ID in RePORTER, you'll get https://reporter.nih.gov/project-details/9273288: Towards the bottom of the page, you'll find a list of publications (that the NIH was able to find) that used the UV300. With funding in FY17, the machine's past end-of-life and the funding's ended (it was one-time funding, with a 5-year (I think) lifespan), but publications are still coming in.
One thing that's worth noting: The grant had no indirect costs. The money went straight to SGI (they were acquired by Hewlett Packard Enterprise in November 2016, and were still being absorbed). The UV300 is not a "set it and forget it" machine. It's not 'cattle'. So, where did the money come from to run the UV300, and answer questions of researchers who (for example) have never heard of "NUMA"?
The money to pay the supporting folks (sysadmins, etc.) came, indirectly, from the researchers: The S10 grant was structured such that the majority of time on the UV300 was for folks doing work under NIH grants. The indirect portion of those NIH grants is what helped fund the sysadmins (like me) who kept (and still keep) the UV300 up and running.
I think grant-funded research is supported enough if…
1. The facilities researchers use (buildings, equipment, network, compute, etc.) are available enough (they are online for some % of the year)
2. Researchers are able to get technical assistance (for the buildings, equipment, network, compute, etc. that they use) in a reasonable amount of time.
What values should be used for "available enough" and "reasonable amount of time"? Honestly, I'm not sure, and since I've got 'skin in the game' my guesses would probably be too generous towards me. Plus, Goodhart's law applies here.