Something or some collection of people are doing something wrong, somewhere in the chain. So isn't this just yet another way to further entrench the modern state of meh-ness in performant UI?
Was insert-startup-product-built-on-.NET fast 15 years ago, for example?
Not trying to be controversial, just wondering.
I agree that a product built by clueless developers stringing together random libraries or bad custom code in React would create slow products.
In your spirit, I might argue that thinking "reactively" and client-centric tends to lead to unneeded requests.
Developing without meaningful API contracts or basic CS skills also makes it easy to unintendedly blow up network payloads.
Caching is hard though.
Have you worked on bad PHP applications that reload on every interaction without caching?
So apps without network latency are faster? Huh.
That sounds like blindly throwing money at the software architecture problem you created for yourself. Supposedly SPAs became popular because your line of reasoning was embarrassingly absurd, in the sense that you do not mitigate the penalty of a network call by microoptimizing the cost of a network call.
Microbenchmark to compare with typedload (which I wrote) and apischema https://ltworf.github.io/typedload/performance.html
Note that pydantic and apischema use .so files, while typedload is just pure python code.
