WASM blobs for programs like these can easily turn into megabytes of difficult to compress binary blobs once transitive dependencies start getting pulled in. That can mean seconds of extra load time to generate an image that can be represented by maybe a kilobyte in size.
Not a bad idea for an internal office network where every computer is hooked up with a gigabit or better, but not great for cloud hosted web applications.
The fastest code in the article has an average latency of 14 ms, benchmarking against localhost. On my computer, "ping localhost" has an average latency of 20 µs. I don't have a lot of experience writing network services, but those numbers sound CPU bound to me.
This would entail zero network hops, probably 100,000+ QRs per second.
IF it is 100,000+ QRs per second, isn't most of the thing we're measuring here dominated by network calls?