This happens, if power supply decoupling is not adequate. Due to very low equivalent series resistance of 100nF ceramic capacitors, enormeous spike currents can flow during the charging phase. So, capacitance of power supply decoupling measures must at least equal the storage capacitance of MAX232 derivative. Perfectly suited for power supply decoupling are ceramic high caps, like 1...10µF/X5R. These are capable to deliver the according spike currents, so that voltage regulator will no longer shut down.

When tantals are used for the storage capacitors, then no relevant problem should occur, because equivalent series resistance of tantals is more than 10 times higher compared to ceramics. But nevertheless, there should be enough power supply decoupling capacitance, at least equaling the total storage capacitance. Otherwise lots of noise is injected into the supply voltage.

Another option is to add a small ferrite bead or resistor into the supply voltage connection to MAX232 derivative. This will also keep the spike currents to a sane level. This series impedance in combination with local power supply decoupling capacitor form a very effective low pass filter to reduce noise on Vcc, by the way.

Kai