I'm not sure this is actually Decentralized. I have been playing with IPFS quite a bit recently, and this looks to host the content on one node. That node goes away, your content goes away. You need more than one nodes 'pinning' the same content in order to actually be decentralized, and as far as I can tell this project doesn't provide that functionality.
That has been my biggest drawback so far with IPFS, there isn't really an easy way to get other nodes to pin the correct content, without passing around big nasty hashes (e.g., QmYwAPJzv5CZsnA625s3Xf2nemtYgPpHdWEz79ojWnPbdG) by hand.
The Dat project, which is somewhat similar to IPFS, and the people behind Beaker have addressed these things. So you can have a site like dat://9900f9aad4d6e79e0beb1c46333852b99829e4dfcdfa9b690eeeab3c367c1b9a/ or you can access the same thing as dat://fritter.hashbase.io using DNS-over-HTTPS for Dat (which is a mouthful, so I've been calling it DSN [Dat Short Names] instead). TTL is controlled by the publisher, just like with DNS. The main downside with the current incarnation is that while the content network itself is decentralized, short name resolution is not, since it's bootstrapped from the traditional DNS/HTTPS infrastructure for now. Which means the short name works only so long as the publisher continues responding.
Beaker is a web browser for Dat sites. The 0.8 release is supposed to happen sometime in the next month, I think. The same team set up hashbase.io to make it trivial to create a short name for a Dat site, and so that you can have Hashbase act as a fallback superpeer/permaseed for your content.
If I understand it correctly, it is decentralized. When pages are accessed they are automatically copied and hosted by nodes. Eventually, garbage collection will delete the copy. Pinning prevents it from being deleted so that the node hosts it permanently.
As far as I'm aware the 'automatic copying and hosting' currently lasts for around 30 minutes. You need to manually pin anything you want to live for longer than a day for sure.
Additionally, this project utilizes IPNS, which clears it's DHT of entries that haven't refreshed within the past 24 hours. So you DEFINITELY need at least one node online, pretty much always, for this content to load.
That has been my biggest drawback so far with IPFS, there isn't really an easy way to get other nodes to pin the correct content, without passing around big nasty hashes (e.g., QmYwAPJzv5CZsnA625s3Xf2nemtYgPpHdWEz79ojWnPbdG) by hand.