Perl has two ways to represent strings,

• without UTF-8 flag as "octet streams" i.e. a list of bytes
• with UTF-8 flag as "characters" in the internal representation°

\x{FEFF} represents the unicode character with the code-point #FEFF, since Devel::Peek shows that the flag is missing, this character can't be found in the octet stream while replacing.

You need to tell Perl how to interpret the read data, the fact that it's "bytewise utf-8" alone doesn't help to see it as list of characters.

The use utf8; in my example just told Perl to read the script's source and all embedded literal strings as utf8.

see Encode for more.

°) which is almost UTF-8, hence the flag is - for historical reasons - a bit of a misnomer