However, we could parallelise Unicode normalisation for very large directories:
Speeding this up would also avoid that entries expire too fast.
