If the second argument to a piped open contains shell metacharacters, perl fork()s, then exec()s a shell to decode the metacharacters and eventually run the desired program. If the program couldn't be run, it's the shell that gets the message, not Perl. All your Perl program can find out is whether the shell itself could be successfully started. You can still capture the shell's STDERR and check it for error messages. See "How can I capture STDERR from an external command?" elsewhere in this document, or use the IPC::Open3 module.
If there are no shell metacharacters in the argument of open, Perl runs the command directly, without using the shell, and can correctly report whether the command started.