I think it depends on the codebase. There are some reflection calls that you can make that can cause dead code elimination to fail, thought I believe it's less easy to run into than it was a few years ago. One common dependency, at least in my line of work, is the Kubernetes API and it manages to both be gigantic and trigger this edge case (last I looked), so yeah, the binaries end up pretty big.

Another thing that people run into is big binaries = slow container startup times. This time is mostly spent in gzip. If you use Zstandard layers instead of gzip layers, startup time is improved. gzip decompression is actually very slow, and the OCI spec no longer mandates it.