We are very aware that new cache key strategy in 2.0 creates some painful situations. However, we are also generally hesitant to do anything that can create non-deterministic behavior, and purging caches can create partial caches and, more acutely, race-condition-style situations where running jobs may run into issues on keys being purged. To do this correctly we'd want some way to queue a request to purge, wait for all jobs that might be using those caches to finish, then block new jobs while we purge -- such machinery is relatively expensive to make Good compared to the value it creates, so it hasn't become a top priority. That said, we are looking into ways to make caching more friendly across the board, and this is on the list of things we get requests for.