They are like seat belts, or audit logs. You don't need so many of them, right up until you do.
OMG so many people died, we need to spare no expense in insuring this doesn't happen again.
[time passes]
OMG these emergency services are so expensive, why do we even have them.
[budget cuts, size cuts, personnel cuts, cuts cuts cuts]
OMG, I'm on fire, everything is on fire, why did the state do this to me, we need to spare no expense in insuring this never happens again.
On a fundamental level, efficiency and resilience are always a trade-off. A perfectly efficient system also tends to be perfectly brittle! In practice everything needs some buffer.
We shouldn't aim for "as close to 100% efficiency on <metric> as possible," but rather "lowest risk-adjusted expected cost, including mishaps." The latter correctly prices rare events.
Yeah. That makes sense a lot of the time. It also makes sense to reduce capacity when the need lessens while still maintaining adequate slack.
Yes, that's why the phrase "when the need lessens" is important.
Besides that, it may be a reasonable decision that 95% utilization is acceptable in order to spend the additional money on something that provides more benefit.