We have generated calibration curves for three of these accelerometers. The sensitivity was found to be 32+/-1 mV/g for all three axes of all three accelerometers. This is significantly different than the typical 57 mV/g as listed. I was using a 3.3V excitation, which I would expect would increase the sensitivity. Is there any explanation for this?

How did you measure the output? What was input impedance of the "Voltmeter"?

The ADXL326 has a 32k series output resistor. If your Voltage measurement device had an input impedance of ~47k I would expect the "apparent" output to ~32mV/g. So first step is to verify the input impedance of your voltage measurement device (generally the higher the better).