Neither deno nor bun do tree shaking because they can’t make the same assumptions as front end code. They do cache the compiled code though, which makes some difference. But the reason S3 SDK is at least 5X slower than bun, is partly due to code size, since JavaScript always have to parse the code on every run. I worked 3 years on cloudflare workers and scripts even 1MB in size are slow and cpu intensive to run (before using it!). Now, even the most common auth for node with basic features is 20MB of scripting. That put a perspective on things like this.