It looks like the constant charging current from the 10k resistor may help a bit to eliminate the undetermined state between fast charge and trickle (or top off), possibly by adding another 1mA to the up and down charging current from the unstable CC pin. But if the batteries happen to be at just the wrong voltage when charging is initiated, it does cause the bq2002T to stay in the oscillating state for some time. I'm not sure why increasing or decreasing the input voltage would affect anything. I already have 15 volts to the LM317 and even with several diode drops this should have plenty of clearance for the 9 volt batteries.
Thanks for you help.