Amazon strongly implies that your data is stored triple redundantly. You'll have to buy more than one disk to get triple redundancy. Also, you'll need to "scrub" the data regularly to detect bit rot (100 bytes per TB are expected to go bad every year). And probably store the files with redundant coding to protect against this bit rot.

Plug: I am working on a startup (submitted to this YC round!) to solve this problem using user^Wcustomer-owned hardware for precisely the sort of reasons you describe. I am looking for co-founders. If anybody wants to talk, email is in my profile.

Plug #2: I wrote and submitted this article about Glacier yesterday but it sank fast: http://psranga.github.com/articles/possible-architecture-of-... Email me if you want to talk about going up against an 800 lb gorilla. :)