The one without any protection is really bad.
The other two have a major flaw, both are asymmetrical. Remember, the sensor has a capacitive output.
This will make the signal analysis a nightmare, let's take the first example with a short 2V peak to peak sinusoidal burst. Before the burst the voltage is zero, then a sinusoidal signal with the lower peak aligned to -0.6V (the forward voltage of the zenner diode) then after the burst a 1.4V (the average voltage of the output signal) slowly decreasing to zero signal.
You can see in the picture a 30Hz 1.5V 20ms burst (VI1) and the output after the clipper (VO1) The zenner diode forward conducting is raises the output signal adding a DC bias.
The other two are a 50Hz 300mV 20ms burst and the output after the clipper, identical but also with negative values that ADC cannot read.
Even with a lower than 0.6V signal, you can read only the positive voltages which is a drawback for asymmetrical shakings like a bump.
I would use a setup like this with two clipping diodes and two resistors that add a constant DC bias to make the entire range readable by ADC:
simulate this circuit – Schematic created using CircuitLab