1: iSCSI can be seen as a USB drive over a network. you plug in, your machine sees it as a drive, you write data. you can unplug, and then plug in somewhere else, and as long as the file system is readable on the new machine, you can get your files.
2: the appliance AWS are offering gives you iSCSI volumes, backed by DAS or SAN storage locally, but also backed by S3 storage in AWS.
So, basically, its like having a drive, automagically backed up to S3, but S3 does not need to know anything about what is on the drive. it could be VMs, Videos, your mail server... anything really.
It sounds to me like enterprisey dropbox: It backs up your files from your local file server to the aws cloud, and if your server dies, bung in a new one and all your files will reappear.
Great idea, although maybe something you could already do with dropbox?
As far as I understood the product page and Werner Vogels' blog post[1] it does not run on file level, but on filesystem level. So you won't be able to access single files from within S3, but rather have whole disk images stored in Amazon S3, ready to be restored back to your local datacenter or to be mounted on EC2 servers.
Actually even lower level than that. iSCSI just makes a block device available over a TCP/IP network. You can use that block device however you like, write random data to it, partition it in to multiple volumes, use it as part of a volume group or disk pool, etc. The individual block device doesn't need to have a filesystem on it.
Except you can use it to power on EC2 machines with those data, so you can use it for dealing with requests spikes, for offsite data elaboration or even Disaster Recovery of your onpremises infrastructure on Amazon.