It was never quite fixed, it seems. There are two standing issues with the Dynamic Classloader
- if you delete and drop in a new jar, it's possible that the delete only triggers, causing any classes in the new jar to fail to load unless the file gets renamed in the ext directory
- if you rename the file in the ext directory, it's still possible that the class files fail to load
I'm speculating there are race conditions around the Monitor being used to watch the file system, but that is just speculation.