Byte counting is one thing, but have you seen the size of AROS 68k executables?
There are examples of commands in the c: directory where the OS3.1 version is a few hundred bytes, yet the AROS68k version is multiple thousands of bytes.
That's the type of thing that makes me facepalm when it comes to AROS.
It's not an issue of C vs. assembly either, as I have been working on real native versions based on their C code that are as small and sometimes smaller than the OS3.1 versions.
Sometimes it's added AROS specific calls, sometimes sloppy linking, often just awkward logic, but it's wasteful.
You can't throw away that much memory and disk bandwidth and expect it to run as well as the original OS did on the same hardware.