Multiple nfs-provisioner instances can have the same name, i.e. the same value for the provisioner argument. They will watch for the same class of claims. When a claim is added, they will race to acquire a lock on it, and only the winner may actually attempt to provision a volume while the others must wait for success or failure. By default, the winner has up to 30 seconds to succeed or fail to provision a volume, after which the other provisioners again race for the lock. This minimizes the number of calls to Provision.
Multiple nfs-provisioner with different names can be running at the same time. They won't conflict because they'll try to provision storage for their own classes of claims.
Given that multiple instances can have the same name, to scale up or down a set of provisioner pods, you simply create or delete pods with the same provisioner name. This can mean adding nodes/pods to a DaemonSet or creating more Deployments, as described in the Deployment doc. Each additional instance would back its PVs with different storage, effectively creating & adding to a "pool" of storage.