Yes, but what does it mean to have a relay? If you mean a full relay, where users can connect to you, great, but I imagine it would have to be through Tor. Also, keep in mind the size, hundreds of thousands of daily events take up a lot of space. However, if you mean a personal relay, where you are only connected to your relay and it bridges the rest of the relays, wouldn't the result be the same?
Discussion
If we assume x number of events per day on the network, the more relays, the fewer events each relay needs to store. Text events are tiny, you can fit all of wikipedia's text on a DVD. If we integrate with other P2P networks like #IPFS, #Freenet, #Hyphanet, etc for storage of media, then space isn't much of a concern. Tor isn't needed, NAT punching protocols exist and are mature, plus IPv6.
Think of it this way: Can your phone store all the text from all your social media posts? What about all of your 10 closest friends? Yes? Then this can scale.
But you don't need to take my word for it, look at existing P2P technologies which have been doing this kind of routing for years. There's no need to centralize around relays.
This way it seems very feasible to me. I am not familiar with NAT punching protocols. Also, if we think about it we could save a lot of requests to other relays and thus make them smaller. It makes sense as you put it.
If as a developer I can help in some project like this it would be great. Since having your own relay would make custom filtering of information much more feasible. I'm not talking about a filtering based on likes like traditional social networks, but a fully customized and configurable filtering.
Nostr is pretty far along and already well committed to the whole "every user is not a relay" paradigm. Which is fine, it's probably "decentralized enough". One of the key pressure points coming up is going to be the cost of hosting media for relays and the tendency for relays to centralize a bit as the average relay hosting cost increases due to increased users and particularly increased media. If we integrate a P2P way to store and to distribute media now, we can avoid much of that.
For example, we have a NIP for associating a pubkey with a LNURL (lightning address). We can also have a NIP for defining a link to an "external media" which is downloaded via Freenet, Hyphanet, IPFS, etc. Let the client do the work of downloading and re-sharing media content while the relays are responsible for text and links to media. At least for Freenet, you talk to it via your web browser via a localhost port. So a nostr client can easily make links to, upload, and download through it without ever needing to leave the browser. The only thing that needs to be true for that to work is for the user to run the freenet client in the background. Unfortunately, they don't have a client for Android or iPhone, but proxy services (like currently used for media) can help with that. IPFS may be another route, I am less familiar with it.
If you're doing nostr dev, this could be something that I think would be on the easier side to implement. I'd gladly contribute to a bounties for it, this is something I'm passionate about.
If I ever start such a project, I will not hesitate to discuss it with you.