RNI is infinitely scalable, given appropriate hardware. For a typical index of about 10 million names, latency is very low and throughput is quite high.
On the high end, for an extremely large index having 100 million names, running on a single enterprise server having 16 cores (2.7 GHz Xeon), latency is less than 3 seconds.
The most common strategy for scaling up is to index all names onto a large multi-core server, and to put multiple clones of this server behind a load-balancer.