Also, backing up, efficiency is not the only metric we should care about. We learned the hard way at the start of the pandemic that our efficient supply chains were fragile due to lack of redundancy. A government agency may benefit from continuously staffing "redundant" roles in order to have capacity available when unforeseen events occur! Nobody needs FEMA until a natural disaster occurs.
It also boils down to the fact that it's not immediately clear what "efficiency" actually means in this context.
Like, I'm a software person so forgive the analogies, but even within software, I can think of three basic (but related) forms of efficiency: execution speed, energy, and memory. Sometimes something will be extremely fast and CPU efficient but at the cost of a ton of memory.
"Efficiency" is a term that is relative to the thing that you are measuring, and doesn't actually make much sense without that.
To your point, having some redundancy can be efficient, if you're optimizing for uptime. This applies to computers and people, I think.
Institutional knowledge often plays a huge role here, but can be difficult to measure. A team member whose experience allows them to quickly provide needed context that would otherwise take hours/days/weeks to obtain might significantly boost their team's productivity without generating any metrics that demonstrate their contributions.
