I've just noticed that my double-percent-encoding example seems to have been double-decoded somewhere along the way. Let's see how this goes:

1. &
2. &
3. &
4. %26
5. %2526
6. %252526
7. %25252526
8. %2525252526