Avoid doing the expensive recursive statistics calculation twice.

This is where most of the KMail/Kontact CPU time during a mail check is
lost, caching accumulated stats per node would be a real solution, but
this change already saves 30-50%.