The OP is describing Conway's Law, and shows that half the battle is being able to define the problem space in a way that minimises communication and maximises utility. As such, it could be inferred that beginning with a monolithic organisation, and splitting off a new team at the appropriate time when a useful problem space had been defined could be a possible best practice, assuming the organisation was flexible enough.

Food for thought, at least.