How this functions by and large is that large content (namely videos and images) are sent to IPFS, text data can also be sent and stored in IPFS as well. So the content gets sent to IPFS and IPFS gives Steemit a hash to reference. I'm not entirely sure about all of the backend of Steemit and what all is offloaded to be frank.

So when you go to a blog post or comment (with images), that content has a hash inside it that reads from the IPFS network and displays that content seamlessly for you in your browser.

I can't vouch if Steemit directly uses IPFS itself, however Steemit does benefit largely from the IPFS network in regards to the likes of dtube and dsound for sake of example.

Referencing po.et wouldn't be much benefit in and of itself here as Steemit could just replicate the same thing on their end with little extra to sacrifice towards it. It would be better to add canonical linkage for those of us who post offsite from WordPress for example.

So in short, it's basically referencing a series of hashes to provide you the content that you consume, namely the larger content.