Since inserting into hashtable may fail, this would result in lost
records. I refactored the code a bit and hashtable node split is now
done recursively, i.e. it divides nodes until it fits.