To find out the truth, I loaded my news.dat into EPM. It turned out
to be 241768 lines in all. With the search function of EPM, it was easy
to locate articles I foud with Yarn. It turned out that the stored mail
I had been seeing was stored in psevdo-newsgroups (like list.Yarn).
So mail filtered to a psevdo-newsgroup is stored in news.dat, while
ordinary folders are not.
It was not possible to see a pattern in the free spaces of the file,
except that there did not seem to be much of it. Maybe I will finish
the database manager for Yarn I started on last year to find out
more about it.