I think the marketing figures offering more than five or six nines of durability are just deceptive. There's no way Amazon or anyone else can model risk well enough to offer eleven nines of durability.
Assuming the real percentage is much lower, objects have already been lost (of 2T total in April 2013). But given the high write to read ratio, the owners may never discover this.
Subjectively some fields either did not provide enough information, e.g., most providers have free ingress traffic, or I could not find enough information about a field, e.g., number of replicas for several providers. I tried to pick a subset of data that was both useful and complete for the defaults; other fields can be selectively enabled.
Glacier definitely has a cost advantage for archiving, although Google DRA offers a compelling price/ease of use tradeoff. Day-to-day storage is less clear; both AWS S3 and Azure have competitive offerings and Google is rapidly improving. Some of the other providers have advantages that a tabular comparison does not capture, for example Joyent offers a fused compute and storage solution.
