After getting your suggestions on using SET REFRESH and FLUSH to make sure shared data is stored, I did some changes. One was to remove a Tablerevert() from a form.destroy event.

I got a backup from the lab last night. Seems to have all records, now 907 more with kornID 12737 to 13645. But at 13521 and 13199 there is a record with only binary zeros. It is obviously written over the record some time after the record was added witn a kornID. This has happened before, but not for about 2 weeks. There has been about 10 such records destroying data.

I have not been able to check the NT write-cache and SP status, but after this I think I will add another server and do updates to another shared table in parallell.