There seems to be a misunderstanding. Yggdrasil is not a single network, it's a tool to make such networks. So if you have some computers with some ability to connect them, you can set them to peer in Yggdrasil, and instantly get an auto-configured irtual IPv6 network. (You can even slap some IPv4-to-IPv6 NAT over it, but we shouldn't mention crimes against humanity.)

It seems to me that the main goal of the project is automatic routing that promises that in any peer graph any peer can reach any other at least somehow, which is fine for many use cases. Only in case of moving big data or very different cost of links you might want to check the node hierarchy and set manual peering from X to Y and Y to Z to make traffic flow the way you want.

Public network is a volunteer-run test ground for real life usage, and a way to connect systems that can't connect directly because of NAT. Internal speed test that is 8 hops away on the other side of the world gives me 30 Mbps, closer one, in 5 hops, reaches 90 Mbps (and the local interface is 100 Mbps), so it has some throughput for casual users.

As far as I can understand, current latency-based routing selection is just the most simple way chosen to optimise the network across the whole world, and prevent random 200 ms hops back and forth to intermediate nodes. If traffic from nodes in one half of the tree not having direct connections to the other half goes through the central nodes, it seems that at the moment it's enough for owners of central nodes to peer with each other, and have enough bandwidth. In the future they might need to take the proportions of available channels into account.