This is only sharing the data within a single mail. It could
be optimized even further by reusing the same instance for every
thread. But I think that's not easy without implementing threading
on the server side.
This decreases peak memory consumption by ~100MB on my mails, when
opening the LKML folder.
Depends on new KMime API (cf. https://phabricator.kde.org/D7920).