Yes, as @Sayak1896 mentioned, node classification currently influences what runs on a given pega node for functionality the engine initiates (listeners, agents, QP and other background processes).  For service requests each node will service a request if it arrives at that node.  As such you can still use node classification alongside a properly configured load balancer so that a set of nodes only handles certain inbound service requests.  Configuring node classification in that case doesn't impact runtime behavior but will likely make it easier to operate the environment.