OpenAI's view of multi-agent is something like: you get the same work faster, but it does not scale linearly.

Only you can decide if that speed is worth the cost.