(Where a "very long time" is probably on the order of a week.)
We occasionally see BFG tickets where there's a failure on ARM, zSeries, or PowerPC from a very old commit for an issue that has since been resolved. This confuses both the Build Baron and Server engineers because the expectation is that recent failures which are reported are actually recent. I can't imagine a use-case for wanting an Evergreen task which spends over a week in the task queue to have a guarantee that it actually gets run.
It should, however, still be possible to manually schedule an older commit and have it run if the task can be assigned to a machine quickly enough. This is still an expected workflow to manually try and bisect when a failure was introduced if stepback couldn't.
This behavior has already been implemented in
EVG-2343 and EVG-3418 for the case where Evergreen decides itself that a build should be scheduled but doesn't handle the case where a user (e.g. ramon.fernandez) manually scheduled a whole version.