1. As you said, the costs involved. You'd probably need to charge people something like $10 per interaction with a voice actor, plus $2 per minute. This makes for a very limited market.
2. Actors need to be waiting for a request on a wide range of characters since there would not be enough demand for them to consistently play the same character all the time. So, when a player triggers an interaction, there's going to be a delay as the actor is brought up to speed on their role, the person they're interacting with, the world they're in, etc.
3. When players return to a NPC, they'll likely be assigned a different actor since the previous one is either in another call or not working. This kind of ruins the premium interactions since the player remembers the voice and small details of the last conversation, but now the NPC has suddenly changed.
4. You're in the business of matching up adult voice actors with children playing games. You'll need a record of every single conversation because it's only a matter of time before you get reports of inappropriate conversations unfolding.
5. This isn't a long term business. Technology gets better each day at speaking and understanding human voice. It will replace the voice actors and it also solves all of the above issues.