Super-niche use-case: our game studio prototyped a multiplayer horror game where we played with cloning player voices to be able to secretly relay messages to certain players as if it came from one of their team-mates (e.g. "go check out below deck" to split a pair of players up, or "I think Bob is trying to sabotage us" to sew inter-player distrust, etc).

Less-niche use-case: if you use TTS for voice-overs and/or NPC dialogue, there can still be a lot of variance in speech patterns / tone / inflections / etc when using a model where you've just customized parameters for each NPC -- using a voice-clone approach, upon first tests, seems like it might provide more long-term consistency.

Bonus: in a lot of voiced-over (non-J)RPGs, the main character is text-only (intentionally not voiced) because they're often intended to be a self-insert of the player (compared to JRPGs which typically have the player "embody" a more fleshed-out player with their own voice). If you really want to lean into self-insert patterns, you could have a player provide a short sample of their voice at the beginning of the game and use that for generating voice-overs for their player character's dialogue throughout the game.