>I had the impression that INSERT INTO wasn't affecting the header. I think I know why I never had this error before. I didn't put the SET REPROCESS in this system. The default is 0 so if there is a collision, it failed right away by returning this message. I usually put it in all my system but for this one I forgot.

Why don't you try an attemp to lock the file header?
VFP introduced the ability to lock the header with RLOCK("0",SELECT())
and the ability to unlock the header with UNLOCK RECORD 0

I have used this on occassion, and it effectively allows you to control
access to the header.