> The best way to plan for 50 years is to continually have a 1, 5 and 10 year plan.
This is the key point to focus on: you don't actually have a backup plan if you don't regularly access the stored data – until then, it's just hope.
I like the CrashPlan + Dropbox approach but there's a key gap: you need something else to check strong signatures (SHA-256/512) on data to avoid scenarios where a bit flips on one of your computers, the file is updated and you don't notice it for a long time, which is really easy for data which is infrequently accessed or where the client programs typically cover minor flaws (most image/video/audio formats).
My favorite example is a JPEG-2000 file which had corruption in the middle of its tile hierarchy – either a thumbnail or full 1:1 zoom looked fine but something like a 50% zoom would show corruption. You'll probably never find something like that without automated checksums until it's too late to recover a copy from somewhere else.
What I'd really like is something like a mature large-file optimized Git repo where there's a strong, checksumed history and it's immediately obvious if a remote version is out of sync. https://github.com/bup/bup and https://git-annex.branchable.com/ are very interesting.
Archive Team is trying git-annex as a way to back up the Internet Archive globally, which will likely lead to changes helping that tool become solid for general usage: http://archiveteam.org/index.php?title=INTERNETARCHIVE.BAK