Since the microcontroller section is so low current, a small TVS is all I needed.
No, this is not the case nor the reason -- the TVS isn't clamping excess current that's somehow some multiple of the MCU current draw (how could it know?). The transient is defined exclusively by the wiring inductance and the bypass capacitance. Using more capacitance makes it worse, because more peak current is drawn through said inductance -- use a big enough cap (assuming circuit resistances stay proportionally low, which isn't the case) and you can blow any TVS ever made!
Also, clamping the spike causes somewhat more current draw, for a longer time, than the natural ringdown of the LC circuit thus formed -- the excess energy being dissipated in the TVS rather than sloshed around the LC circuit.
How much is necessary? You can calculate from some basic estimates, of course. The peak current will be no more than Vbatt / sqrt(L/C). A meter of hookup wire (twisted pair, not loose!) might be about 1uH, so 10uF bypass connected to a 12V battery will draw a peak no more than 38A. How much energy? 1uH at 38A peak is 722uJ. So, you need one rated for at least 38A peak, ~1mJ or more avalanche energy, and probably an operational rating of 12-15V depending on your supply's tolerance, so that it will clamp around 20-25V, and be suitable for use with 30V devices. Make sure your regulator is rated accordingly, or else provide the necessary compliance by adding more regulator (or converter) stages, or a transistor and zener pre-regulator, or something like that. TVSs are usually divided into families by peak power, rather than peak current, so if you figure the peak voltage is ~20V at 38A, you need a 760W device -- an SMBJ or P6KE class device would be marginal to insufficient, but an SMCJ, 1.5KE or bigger device should last a very long time.
The situation won't actually be as bad, because ESR, DCR and other losses cause some dissipation as the current climbs to its peak value. You can tell roughly how close you are based on the ratio of total ESR (battery, wire, capacitor) to sqrt(L/C): the closer to 1 the ratio is, the less it rings, until for ESR > sqrt(L/C), it doesn't ring at all (and more resembles an RC charging curve instead).
The waveform in a previous post is a perfect example of things that work against you: nonlinear (any ceramic but C0G) dielectrics lose capacity strongly with voltage, so the inductor current causes the voltage to just keep shooting up and up and up! (If C remains constant, the overshoot cannot be worse than 100% -- just as the classic physics demo of swinging a pendulum at ones' nose.) This might be another good reason to use a TVS, but also a good reason to use a larger, lossier capacitor. An aluminum electrolytic -- 10uF 35V might have around 5 ohms ESR, which is well above the resonant impedance of ~0.32 ohms, so it will charge slowly. If there's a few 0.1uF ceramics on the circuit as well, they will want to resonate with a few ohms impedance, which is well damped by the ESR in parallel with them -- so it's great for both overshoot and damping circuit parasitics!
Tim