Tom,
>I would classify bugs as "by design". In that case nothing should be done about them. We must follow the logic. :)
I'm saying two things might be by design. scoff at this all you want, I really don't care especially at 5:30 on a Friday whan I should already be gone for the weekend...
1) the 32kb difference is probably by design from deleted index headers after the reconstruction. And this is a pure guess on my part, I really am not interested enough in this aspect of the problem to fire up hexedit() to investigate it.
2) if the new node splitting is causing the file size to increase, that might indeed be by design to have faster inserts occur during the more normal use of the table.