Our Linux file server uses LVM on top of RAID-5 (4 120GB SATA and PATA drives, resulting in 240GB of space with 1 drive in hot standby). We do a full backup to (SCSI based) DDS-3 tape (using LVM snapshots) every Sunday. We then do incrementals each day to DDS-3 tape. Data is stored using standard tar format (so there is NO chance of NOT being able to access the data in 5, 10 or even 50 years). We also do not use DDS-3 compression, so there is no tying the data on the tape to a specific drive format (we currently have both Sony and Seagate drives). Our critical data requires 6 12GB DDS-3 tapes (72GB) - we are currently looking at upgrading to DDS-4 (24GB per tape so only 3 tapes needed).
Each weeks worth of tapes are then stored in a safe. The first backup of each month is then transfered to our safety deposit box at a local bank. After six months or so we purge every second monthly backup set (just re-use the tapes). After a year we perform the same operation to yield 3 monthly backup sets per year archived forever. We have been doing this since 1997 (back with only DDS-2), so you can imagine that we have quite a few tapes.
Although tape is not by any means a perfect medium (it is susceptible to strong magnetic fields and tape tension can be a problem) it has proven itself to the IT world for decades to be very reliable. We plan on staying with DDS based technology, which is backwards compatible (at least for reading). If for some reason DDS is no longer a viable technology we will have to re-visit our archived data (at least one snapshot per year) and transfer it to a new medium.
Why do we keep data for so long? Well, I have personally lost files that were supposed to be on a hard-drive which ended up being corrupted. Our so-called backup was using an unreliable medium, which had degraded. There was only one backup made so we lost the data. This system was designed to avoid such a problem, yet nothing is perfect.
Where loved ones are remembered: Memoriam.org