Just came across an interesting utility called JungleDisk. It creates a virtual disk on the desktop which is automatically encrypted and backed up to Amazon's S3 service. The disk uses local storage as cache for fast local use, and is accessed via WebDAV - which is how it can figure out what to send to S3 in the background. It only sends changes, AFAICT. This looks like an eminently practical solution - even for larger datasets, such as a personal photo collection.

Haven't used it - beyond a quick tryout. JD only caches files up to a (configurable) limit, the rest gets pulled from S3. Also, it looks like offline mode is not supported (yet, apparently) - though setting the cache big enough to always hold everything might do the trick.