That's a 24-hour-news-network-pundit grade straw man he's created there. The replication strategy used by Cassandra is sourced from the Amazon Dynamo design (shared with Riak and Voldemort) and is intentional. It's not a bug; it's the way it's supposed to operate. There are many large scale distributed systems built on this replication strategy that have excellent durability track records -- Amazon S3 is one of them.
The back pressure argument is laughable. In Cassandra, a write consistency of greater than ONE will ensure the data is replicated to more than one node before acknowledging to the client.
You use argument from authority ("It's based on Dynamo, which is made by Amazon, so you know it's good!") while accusing me of a strawman for reporting my own personal experiences.
Good thing there is no god, or you should have to watch out for lightning.
Actually, this is empirical evidence of the validity of this approach gained through observation of successful, real-world implementations. I'm not saying it's good purely because it's from Amazon or because any one person said it was good, but that there are many successful large-scale systems built using these principles that do not suffer the durability issues from your hypothesis.
While not proof in the scientific sense, this type of evidence is used as a basis for scientific theories that explain much of what we know about the world.
However, I still look forward to your detailed, rigorous rebuttal of my arguments in the previous comment.
The back pressure argument is laughable. In Cassandra, a write consistency of greater than ONE will ensure the data is replicated to more than one node before acknowledging to the client.