UNIX for Advanced & Expert Users

I'm trying to do a incremental backup for a big NFS. Since space is not an issue, I don't want to compress them or end up with a big tarball for full backup( and a series of small tarballs for incremental backup). Basically I want the TAR backup/restore functionality but not TAR files up.
Requirements:
1. files/directories are transferred directly so that I can browse through the backup easily.
2. Have a snapshot file that can be used to do incremental backup. (like the one Tar generates, which used to keep track of changes)
Any suggestions? Thanks!

Note: Rsync is not an option, because target filesystem is not unix-like. It only supports GET/LIST/PUT.

Just me... I'd use Linux as the holder of the backup data. rsync, as mentioned is the best solution, but in lieu of that, sounds like you'll have to write a screen scraping (?) ftp like manager to loop through and grab everything. Could mean multiple passes. Might help to know the target system? You'd be surprised at what people have experience with.

I'd use Linux, because assuming you can roll through and pull the whole tree across ala ftp style (maybe you're system supports scripting on xfer??).... then you can use the GNU cp -l option to create a link shadow of the backed up area... this is a cheap mans snapshotting. So as long as the backup and shapshots on the receiving system are within the SAME filesystem, you can use this cheap snapshot mechanism to keep track of multiple days... AND you can browse through it all.

What are the two Operating Systems involved? Please be very very specific.

How far apart (in network terms) are the computers? Are there any stoppers such as firewalls or restricted bandwidth?

What network file transfer software do you have available? Obviously GET/LIST/PUT implies ftp but with better knowledge of your systems there may be more?

If we are confined to ftp, do you have enough disc space to copy back an archive when you want individual files or do you expect the target system to deal with such requests?

How big in measurable units is "big" ?

I'm backup things from NFS on RHEL5 to S3, but I cant use all the features that S3 provides(specifically versioning), because I need to go through a mandatory encryption.

bandwidth is not an issue.

I dont have space to copy back a full archive, but I am considering splitting them into small parts.

data is about 5TB