I wonder if this is because one never rounds down to zero, so that finer grained estimates automatically pad themselves. I've noticed that estimates run higher and higher, the longer the planning meeting lasts.