The same queuing problem exists all over in computing as well and is similarly poorly addressed often as not. Japanese car companies and more generally lean engineering is all about fixing this. I highly doubt worker assignments are anywhere near optimal and are probably mostly whatever the job foreman feels like is right.

Some groups have this down to a science for their particular industry (say, Apple supply chain logistics) and others aren't even aware of the issue and pass it off as expected.

It would be an interesting study to get queuing theory people, lean experts, or some such qualified group to look into public construction and maintenance projects to simulate and analyze optimal vs. actual. No question this would be very hard to do because unions hate this kind of thing and people don't like being measured over their shoulders and the basic implication that they're "lazy" or inefficient or whatever. (it's hard to measure and improve worker output while actually making the worker happier too because there tends to be a Stanford Prison Experiment vibe where you get managers nitpicking lunch breaks and the basic pleasantries of working instead of bulk errors like allocating skill sets)