It's not as lop-sided as this article might have you believe and has largely been mitigated as of late. This is because Cassandra uses read repair, which is a big component of it's strategy to make both reads and writes to scale linearly while also ensuring durability.

What is your suggestion otherwise? Any distributed database that is going to be inexpensive, performant, scalable, and durable will need to use some kind of quorum read repair system. Riak, Voldemort, and Dynamo all use read repair with high levels of production success.