PARIS may have a typical length in Morse code for words in English, but that wouldn't hold in other languages, you're right about that. I assume someone took some text samples of English (perhaps even samples of typical plaintext messages sent by Morse code) and calculated the average length of the words in Morse, and then found the word PARIS to be very close to this length.
But CODEX is language-agnostic, assuming we're still talking about the basic 26-character version of the Latin alphabet + the 10 numbers. This is because that one was mentioned to be the standard for groups of 5 random characters (either 26 from the alphabet or 36 alphanumeric ones). If you're transmitting code groups like that, it's likely a plaintext message that has been encrypted into 5-character groups, those are transmitted, then decrypted again by the receiver of the message (you'll have agreed beforehand what the encryption keys are, and the encryption algorithm). It doesn't matter what the language of the plaintext is, that just necessitates making new algorithms that convert the text to those same 36 characters (a-z and 0-9).
For CODEX, I assume someone calculated the average character length for those 36 characters, then found a 5-letter word where the average character length was about the same as that for the whole character set. As for "why a 5-character word", I assume that one is to match the 5-character word of PARIS, and/or to match the random 5-character groups used when sending encoded messages.