You might want to briefly overview the gateway model in particular to associate the terminology used below - in particular, FLIP, ADVERTISE and PULL. The gateway and associated components are building blocks that help support various use cases on top. Some ideas that are possible with these blocks:
Provide a centralised workspace for introspection, but distribute your actual communications. This avoids a bottleneck with a fully centralised system, whilst allowing you the benefits working in a centralised way.
Think of web services, but on the lan. You might have a computer off in one corner of the lan that is always up and you'd like to expose it as an external computation unit for your ros robots.
The key points here, is that you don't want to have to start it and stop it with each robot - it just runs 24/7. So it needs its own master and a way to communicate with the robot masters.
Robot Access Point
Robots are typically wirelessly connected to the rest of your lan and you might want to collect information about the robots at a single access point on the hardwired ethernet backbone of the lan that can be easily utilised by other components on that backbone. This avoids the complexity of distributed systems, but does allow a practical means of easily controlling the data flow where it's useful.