I wouldn't call it "covered". What is says is: "If randomly distributed, the probability of the same redundant piece being hosted on the same controlled node is statistically quite small." It does not attempt to quantify what "quite small" means. So I'll try to do that.
Assume that you have a million blocks stored in the network. Each block is stored as three identical copies. Then assume that 1 % of the storage network is taken down at the same time. The probability that all copies of an individual block got lost is one in a million. However, the probability that at least one out of your one million blocks was lost is approximately 1-1/e or 63 %. So in this scenario, you are more likely than not to have lost data.