[this is an informational side note for casual visitors who might read this thread]
To clarify that statement, you can delete things from any ipfs node that you have control over -- unpin the content and then either wait for garbage collection to run or explicitly run it yourself. That will delete the content from your node.
What you can't do is delete stuff off of nodes that you don't have control over.
The classical http web parallel: you can delete content off of your own machines and you can delete content off of any servers that you have control over, but you can't reach into other people's machines and delete the copies they already downloaded. With the classical http web you can't even see that those downloaded copies exist on the network.
To remedy this, ipfs-cluster is designed for situations where you want to run an entire network of coordinated nodes, where a leader is able to decide what is pinned or unpinned (and eventually garbage collected) on any of those nodes.