According to reveman@, locking discardable memory is not necessarily a cheap operation. However, it is cheap compared to a typical image decode. As such, it should be treated as a cheap image decode. However, this means that the lock should still happen on the worker thread instead of the compositor thread.