Rationale: 1. We get a fair number of bug reports for bits like this when they're really intended for ebuild maintainers and aren't of much concern to users, so it's quite noisy; 2. In bug 781860, while the issue there didn't end up being a GLEP 65 check, Zac posted a patch to allow timing GLEP 65 checks. They don't tend to be very significant at all but I only checked on quite a fast machine, on only a couple of packages, and of course the checks can scale worse w/ image size or other factors (or just regress over time). Doing less work is generally a good thing.
The only quirk here is that some of our essential NEEDED detection is part of some of the QA check machinery but it should be ok to disentangle (don't let this put anyone off working on it, just noting it). Ideally we would even do that part in Python to avoid duplication/running it repeatedly.