According to its data sheet, the microphone consumes 2mA when using phantom power (at 48V) and 0.15mA when using battery power (at 1.5V). It also has an output impedance of 250ohm on phantom power, and 350ohm on battery power.
Essentially it is sort of a different microphone on battery power, using very low power. That means that bursts of high signal or a breakdown of the polarisation voltage (because of the membrane touching or large signal) will require longer for reestablishing working conditions: both polarisation voltage as well as preamp voltage will require a moment to reestablish when using step-up converters (and smoothing capacitors) given the very low power draw on the battery of the NTG2.
In contrast, phantom power will likely power both preamp and polarisation voltage fairly directly and drawing quite more power. When recording in the field from a battery-powered computer and audio interface, using a separate battery for the mic and switching off phantom power might make for longer recording times because the efficiency of the NTG2's battery use may be hard to beat.
In short: one could consider this more a sideeffect from a feature rather than a bug.