I will assume TCP/IP over Ethernet. TCP/IP & Ethernet networks are typically segmented based on several factors with the most being user base or end point base. I.E. How many IP end points that need very fast access to one another being the chief concern. Other factors also come into play too... User population, etc... but let's focus on performance or speed.
The reason I focus on speed is because of the statement of "eliminates network diameter as an issue." Network diameter is mostly a speed issue at hand.
The way you achieve the best or fastest possible speed or bandwidth is to minimize the number of devices the traffic must flow through (routers and switches) and lowering the number of devices connected to a device collapsing the end points (switch); that is I'm assuming a star configuration here.
By reducing hops, you eliminate latency and I/O. You also have physical constraints you eliminate such as fiber to copper, etc...
By reducing the number of devices, you reduce switch I/O (ARP lookups)
So the diameter is often the size in terms of end points and devices your endpoints need to talk via.
The common rule of thumb you often see in TCP/IP & Ethernet books is that the faster the network, the smaller the network diameter. Network diameter being the two things mentioned above (hop count, end points per subnet).