I think we totally should do this, but with a really long expiration period (3 months suggested by Brian seem nice).
The concerns raised by a few people here are valid, but I think they could be offset with some simple improvements:
* Never re-abandon a patchset that has been previously restored (note that I mean patchset, not changeset). * Send a notice (a comment) on the patch that it will be abandoned, explaining why we do this and how to prevent it.
How about this?