the data I store isn't really compressible. I'd probably just go with replication factor of 2. And I already have 18 (or 21 if I'm willing to shuck some drives), looking for the second set of 18 (3x12 bay servers).
But I do a lot of clustered computing stuff (databases, docker, kubernetes....) and I tend to dislike paying for the cloud (somewhat because I'm forgetful and would end up with a big bill).