It takes me about 250ms to load hackernews as seen on the browser debug network tab. The total compute for that request is minimal. Speeding this up may reduce total infrastructure costs, where if everything is kept static you will see benefits long term, but in terms of reducing total latency, it will be next to irrelevant.
I can list a ton of tools I use daily that are incredibly slow, but don't even need the network most of the time.
By "incredibly slow" I mean comparing what it does versus how fast it could be on an average machine.
Network and other I/O latency is often used as an excuse and sweeping generalization. It's never that easy. You need to actually look at a running process and gather data about it before you can make claims about what the bottlenecks are. Also even if there are constant factors that you cannot optimize it doesn't absolve a program from being slow in every other aspect, secondly there are techniques to mitigate and isolate latency bottlenecks.