Even excluding all the details of tarsnap's design, using glacier for backups has always seemed totally wrong-headed to me for a simple reason: it disincentivizes checking up on your backups, which is a key part of doing backups. No point waiting until you lose your data to realize that your backups had developed a glitch.
Add any minimal restoring cost and glacier costs about the same as regular storage -- with a lot less convenience. I honestly can't think of a use case for something like glacier, where storage is cheap but reading/writing is expensive.
I would personally be totally fine with a backup solution which is orders of magnitude slower and several times as expensive in the rare restore case if it meant that the typical "write only for long periods of time" scenario was significantly cheaper. My personal backups are not something I'd ever need to restore in a hurry.
The one big downside to the "cheap backup, expensive restore" approach is that it discourages testing your restore.
But for most personal uses, a Glacier-backed variant of tarsnap would be extremely appealing.