The problem isn't one of characters versus bytes. The problem is the definition of character in the context of Unicode text. The scalar reverse function and other built-in string functions operate on Unicode text using a naīve and inadequate definition of character. Pointing this out and offering a workaround is the raison d'ętre of moritz's 2008 tutorial.
The issue of what reverse does when fed, say, the bytes of a JPEG image are utterly irrelevant to this discussion, which is about Unicode text. I don't understand ikegami's insistentence on trying to fold into this discussion unrelated contexts. Your reply dramatizes how ikegami's contrarian non sequitur needlessly confused the simple and self-evident conclusion I made in my post.
Here's what I wrote:
The documentation of Perl's reverse function states: "In scalar context, [the reverse function] ... returns a string value with all characters in the opposite order." But it doesn't, at least not for a sufficiently modern, multilingual, Unicode-conformant definition of "character." It reverses Unicode code points, not characters in the usual, well-understood sense of the word.
One or the other is wrong: the behavior of the reverse function or the reverse function's documentation.
If I understand the design principles of Perl correctly, the reverse function should properly reverse extended grapheme clusters when the thing being reversed is Unicode text (and Perl understands it is Unicode text), and it should reverse bytes otherwise.