But I think having a lot of dead experiment branches is like having a huge back log. It just some place where ideas go to die.
You stumple upon them once in a while and start thinking about why you never finished them and about how this was a good idea but management killed it and so on. I think it is better to delete them and set your mind free so you can focus on the tasks at hand, not the past.
The source repository is a workplace, not a museum.
Just create a `bla-rearchitecting-experiment.md` in your `doc/` and paste the relevant code and a paragraph or two about what it does, why it could have been better and why it was abandoned. Now your history is in the place it should be, and not cluttering up the workspace of your successors.
History belongs, well, in history.
We can (and should) safely delete commented code because we have branches which track the evolution of features and random experiments that might someday be useful. So long as the branches exist, we can always recover that code we just deleted.
But there's not value in deleting old branches. They represent the history of a project's development and should be respected as such.