Nope. The problem isn’t the amount of the data, but how quickly (here: slowly) it can reach its recipient.
Excellent idea. This could indeed get rid of much of the delay, especially if relays could be picked in a way that they are somewhere along the natural path between the two nodes. For example, a number of nodes could be tested, then the best few kept. Fun ideas like these are what I meant when I wrote we need to escape the routing layer 
Things to consider:
- high churn: people walk around, so new relay connections need to be set up between players all the time, maybe even pro-actively
- relays need to know the real-world identity (IP address) of the players to be able to function as relays (obviously), and we have no way to make them keep it secret: basically, we just got rid of all the awesome trickery that keeps the SAFE network safe…
I was thinking aloud in another thread about building a TOR-like scrambling layer built “somewhere in-between” the base IP network and the SAFE network: it would use the SAFE network (on top of the routing, in “XOR space”) to organize itself, but it would communicate actual data directly via IP (crust, “IP address space”).
Now, if the nodes of this scrambling layer are aware of the distance and bandwidth between each other, then they could work out a compromise between security (information hiding) and performance: more or less number of hops, random v.s. directed hops. The different applications could request different kind of connections based on their requirements. This, in effect, would amount to something like a “parameterized TOR network”.
This network would use one-time connection ids: two SAFE nodes could agree on their transient ids, then the scrambling layer would connect them.
If we have a scrambling layer like this, it can be used for any real-time sort of applications, not just gaming. I may make this idea into a thread sometime… 
Fun idea: Something like this could form the base / evolve into an “off-the-grid” transport layer to serve as a replacement for the physical internet. The safe network, in its current form, needs the internet (which is completely okay) but maybe it’s not a bad idea to figure out alternatives. other ways go into a P2P mesh in the physical aspects of the network as well (not as its sole transport, but one of them). A nice circle: this network would be organized using the SAFE network, and it would form the transport (one of the transports) for the SAFE network.