This mission of the delegate consists in using the collected fees to pay for seed nodes and backbone nodes. Higher pay means more nodes can be funded.
Stability of the network is guaranteed, amongst others, via good connectivity. To this end, it can be useful to have specialized nodes that ensure the network is as close as possible to a small-world network. This ensures amongst others that the average distance between any two nodes in the graph is low, which translates into quick broadcasting of the blocks.
Seed nodes are important because they are the initial entry points to the network. As such, they should have a very high uptime and allow lots of connections while facilitating peer exchange.
Backbone nodes are nodes that are interconnected in a fully-connected graph, on top of the other connections they have to the network. If delegates connect to these nodes, they can ensure that a block they broadcast will only require 3 hops at most to reach the next producing delegate.
Backbone nodes do not perform peer exchange in order to hide the IPs of the connected delegates, to protect against potential DDoS. Backbones nodes are public nodes and their IPs are known, though, so nodes (delegates) can easily try to connect to them and maintain a few live connections to backbone nodes.
This acts in fact as a high-speed backbone for the delegates that ensures low-latency when propagating a block. This network should be open to any connections, but should de-incentivize normal nodes to connect to it, or make sure somehow that it stays always available for delegates.
Care should be taken that the previous diagram doesn't turn into an island where only the backbone nodes and some delegates are, which could create forks in the blockchain. A proposed solution is to have the seed nodes connected to the backbone nodes too. Seed nodes see a lot of churn and maintain a very high number of connections and thus should not end up in a minority fork very often. We can't prove mathematically that it makes things better but it somehow feels right :)
To not put all the eggs in the proverbial basket, seed nodes and backbone nodes should be run on different geographical locations and VPS providers. As the backbone nodes are parts of a fully connected graph, a single node being down shouldn't affect connectivity at all. In fact, one would need to take down nearly all of the backbone nodes in order to severely affect the network. Note that this doesn't make the network more centralized and fragile, because if you attack and take down all the nodes you're basically back to the starting point without them. This does make the network more specialized, though, and more efficient in accomplishing certain tasks (namely: fast block propagation and high connectivity).
In fact, if the small-world network constraint can be successfully enforced, then this means that not only the nodes directly connected to the backbone will benefit, but also all the nodes from the network (albeit with a slightly larger average distance between nodes).