Electronic – limiting servo current consumption

Typically, any feedback system requires capacitors, so that small fluctuations in the power supply due to movement of the motors do not create another unintentional feedback path. In this case, the charge lost in 1 discharge cycle of two 250uF capacitors is minuscule compared to your battery capacity: 250 uF * 5 V is about 1.25 mC, by comparison 1 mAh (where most batteries are on the order of thousands of mAh) is 3600 mC. Therefore, a typical 1000mAh battery can theoretically support many million charge and discharge cycles of your caps.

Not having capacitors may very well reduce your battery life, if small fluctuations in the power supply cause noise in your servo feedback readings and therefore cause the motors to move unnecessarily. Powering up the motor (even for a couple of milliseconds) is likely to waste magnitudes more energy than charging and discharging the caps.

You have several things to worry about in this setup. A schematic of what you have exactly would have helped to target the advice more to the exact problem, but here goes:

First of all, you may not necessarily need insanely large capacitance, but you need capacitance with a low ESR (Equivalent Series Resistance). If your 470uF capacitor has a large resistance the regulation of the buck converter will improve and the peak current capacity of the capacitor will increase, solving the problem from two ends. That's not to say you may not need a higher value if you don't want a dip at all.

Secondly, you need a low ESR capacitor on the input of each converter stage as well, to make up for resistance in batteries and/or wiring, so the converter has a local store of energy to pull from when the output suddenly needs 1 or 2 ampere for the motor's cold start.

From the datasheet I expect the response of the regulator to be quite sharp, so it should be searched in the capacitors first. If you cannot get or afford a very low ESR capacitor, you can combine two or three lower value capacitors, as their ESR will be used in parallel and as such the total ESR of the capacitor-group will be much lower.

The second picture does seem to show a retry/restart of the motor and the regulator's response to that transient current. It's likely the servo has a little margin when the power to the motor falls off. More advanced servos are better tuned to prevent from showing behaviour like this.