In practice I don't think something on the order of 10k small files is going to burn the cluster down--I'm on the hadoop team, and I'm familiar enough with what normal cluster usages look like