I think @pz1 has a good point.
The network architecture we're developing may not be the golden solution for all use-cases. If we had to define a solution that is perfect for everyone, we would end up in endless discussions and not get anywhere. Nothing stops you from building a solution that works in a different way. There are already great alternatives, such as @brocaar's loraserver.
We don't know what kind of infrastructure will be needed in the future. We feel that what we're building now is something that will support the current use-cases and is flexible enough to change if needed. And yes, we know that it's not perfect. It will evolve over time, and we encourage you to investigate other options. We'd love to hear your experiences with those options, and will definitely consider them in upcoming versions of the network architecture.