In the case of C2 built daemons you may be right. In the case of a server daemon written in a systems language and using an appropriate tech stack the situation is not nearly as dire.
The *only* thing 10 physical servers give you is redundancy and clustering. Its clustered processing power is for the most part equal to or slightly less than a single server of the combined physical specs.
In terms of multiple software servers (ie, multiple database instances), it's common to have a read only master and readwrite slave setup, and again, can be happily implemented on multiple servers or a single server.
"thousands" is not a big number, and I'd be weary of whatever hosting it is you use if it is. I run all of my servers on rented rack space.
And again, depending on technologies used and technologies available, things like broadcasting and caching can be handled by mostly inexpensive third party services.
I understand C2 users are typically not programmers, so there's a lot of misconceptions and hyperbole.
To be absolutely clear, I am refuting the notion that "100s of users" would require "10+ servers" especially when the concept of server is so ambiguous.
Total bandwidth is literally an issue solely of the pipe leading to and from your datacenter and the hardware that handles it.