TFA suggest a 900% overcapacity, not a few percent. I just skimmed GP's article but it seems to suggest a ~100% overcapacity for streaming-video specifically.
"100% overcapacity" is also known as "50% load factor".
It's common knowledge to run your API nodes at 20%-60% CPU load, no more, exactly to curb tail latency.
just think about your competitive edge if you could manage to cramp more load w/o affecting latency
Depends on what you have to pay (and not just in money) to get there.