When you put a server to work, the more users using it, the more server power will be needed.
Major websites do not buy bigger machines when their audience increases. The concept of load-balancing is having a single entry-point that delegates the requests to the next level. Most commonly, the next level is made of servers but it can be more load-balancers. Load-balancers may be hardware or software (like NGINX).
Neo4j's new protocol handles load balancing. Thus, creating a cluster with more replica servers in it allows more throughput while making the cluster a virtually bigger machine. Additionally, a query can be divided so that its parts are executed in parallel on several machines.