- Non-IP Based. I think that Dotcom refers to the fact that the "webpages" or whatever they're going to serve are not hosted on a specific IP but are instead distributed in the blockchain; and he talks about IP because either a) he doesn't know what he's talking about or b) he's oversimplifying a lot. In any case, the MegaNet would run over IP because no ISP is going to let non-IP protocols go through their networks.
- Using mobile phones. Even when on WiFi, that's going to eat batteries for breakfast: users aren't going to like that. And another issue: MegaNet traffic would still go through traditional ISPs, so they would probably be able to filter it.
Another point is that Android, iOS and Windows Phone have closed, controlled stores. That's going to make it difficult for them to implement.
- Using the blockchain. I don't know why would they use the blockchain, which is a method to ensure integrity over time (e.g., you don't spend two times the same bitcoin). What would would make sense to use P2P webpages, like BitTorrent's Maelstrom does [1]. Maybe they use the blockchain so they can keep a correct history of a certain webpage, who knows. Another option is that they are thinking of using the authentication part of the blockchain, but one could do that with Maelstrom using PGP signed pages, for example.
In any case, they'd face significant issues. For example, they would be introducing useless delays when creating/updates webpages (blockchain relies on proofs of work to avoid malicious nodes to insert duplicate/forged transactions). But if they make the verification process fast, they would be more vulnerable to malicious nodes.
Another problem would be how to store the information they want to transmit. The blockchain works because the information in each transaction is just a few bytes long. However, a webpage is way bigger than that. How do they store not only the current versions of the webpages, but the old ones too? I don't think the combined storage of mobile phones is enough to hold all that information. Maybe they could use deltas, but that's only going to make browsing slower (instead of one block, you'd need to retrieve several to reconstruct the page) and they would not avoid completely the problem: what if the difference between versions is too big? E.g., encrypted pages: deltas would be useless then.
- "New layer of encryption that will run through all digital communications". Well good luck with that.
- "We are going to use very long keys, systems that will not be reverse engineered or cracked by any supercomputer". Long keys imply long encrypt times too. With phones, and probably with several communications with several different nodes. That's going to be interesting.
- "If you have 100 million smartphones that have the MegaNet app installed, we'll have more online storage capacity, bandwidth and calculating power than the top 10 largest websites in the world combined". Distributed computing doesn't work like that. You don't just "add up things". Yeah, you may have terabits per second of bandwidth combined, but using single nodes that maybe can get up to 1Mbps. You have a lot of processing power but have to distribute any computation through a lot of nodes because no one has enough single computing power to do complex tasks fast. And you have 100 million smartphones distributed through the world: good luck with the latency and reliability of the connections.
In any case, I think this is either going to be a Maelstrom clone or a complete bluff.
[1] http://project-maelstrom.bittorrent.com/
Edit: added a few points.