Content Delivery Networks offer good scaling opportunities. In my limited experience, pricing of bandwidth and storage space scales better with S3 than most shared hosting alternatives. Amazon's CDN also takes care of the actual resource scaling for you, whereas shared hosting would max out.

With S3: if nobody turns up, you just pay storage for the PDF (less than pennies, $.14 per GB); if everybody turns up, Amazon handle the extra load and you pay a proportional amount in bandwidth fees.

With shared hosting: if nobody turns up, your hosting fee remains the same size (pounds or dollars); if everybody turns up, the shared host falls over and the hosting company invokes some clause in their contract with you to force you to host with S3 instead.

>Couldn't you do about the same with some cheap shared hosting?

Hit Events, i.e. your item is suddenly a hit, come fast and die fast many times. If you're not scalable, when you do get hit with a publicity event, your infrastructure can melt and then you miss out on a huge majority of sales.

Sure, I understand that I guess. My question is whether the demands on the servers hosting the OP's book ever met the level where S3 was really necessary or not. It doesn't really sound like it did, judging from the numbers.