but, I don't think you can logically argue that the Perl source shouldn't be utf8.

That's not at all what I'm arguing.

I simply made the experience that concatenating binary strings and decoded strings lead to nasty errors. So if you don't bother to decode strings on input, telling Perl that the source file is UTF-8 is harmful.

If your script or module is written in UTF-8, input and output data is UTF-8, but you don't decode or encode, everything "works" in the sense that the output is correct UTF-8 (though of course the strings don't follow proper Unicode semantics internally). In this case use utf8; leads to incorrect output.