The problem is that air gaps not only influence the base capacitance, they also influence the capacitance of the touch sensor when someone actually touches it.

The air gap essentially acts like another capacitor that you put in series with the touch sensor. This additional series capacitor worsens the coupling between the microcontroller (or other sensing circuit) and the actual sensor.

Here's an example: Let's say you have a touch sensor that has 10pF capacitance when it's idle, and 100pF when someone touches it. That's a difference of 10:1, and easily detectable.

Now let's add an air gap with an equivalent series capacitance of 10pF. This means that, in the idle state, you have a 10pF air gap in series with the 10pF of the sensor. That's 5pF when the sensor is idle. When someone touches the sensor, you still have the 10pF air gap, but in series with 100pF from the sensor. 10pF in series with 100pF is about 9pF. Now the difference between the "idle" and "touched" states is only 1.8:1! (9pF when touched, 5pF when idle.) This is much, much harder to detect.

So, in this example, adding an air gap reduced the sensor's sensitivity by more than a factor of 5.

The problem is that air gaps not only influence the base capacitance, they also influence the capacitance of the touch sensor when someone actually touches it.

The air gap essentially acts like another capacitor that you put in series with the touch sensor. This additional series capacitor worsens the coupling between the microcontroller (or other sensing circuit) and the actual sensor.

Here's an example: Let's say you have a touch sensor that has 10pF capacitance when it's idle, and 100pF when someone touches it. That's a difference of 10:1, and easily detectable.

Now let's add an air gap with an equivalent series capacitance of 10pF. This means that, in the idle state, you have a 10pF air gap in series with the 10pF of the sensor. That's 5pF when the sensor is idle. When someone touches the sensor, you still have the 10pF air gap, but in series with 100pF from the sensor. 10pF in series with 100pF is about 9pF. Now the difference between the "idle" and "touched" states is only 1.8:1! (9pF when touched, 5pF when idle.) This is much, much harder to detect.

So, in this example, adding an air gap reduced the sensor's sensitivity by more than a factor of 5.

This applies to mutual-capacitance touch sensors as well: If you have an air gap in one of these, you'll get a "gap-plates-gap" arrangement, which makes the problem even worse as there's now two air gaps in series that reduce the sensitivity.

Let's run the numbers for an example mutual-capacitance sensor as well. We'll use the same numbers: 10pF mutual capacitance when not touched, 100pF when touched. (10:1 difference)

Again, we'll assume that the air gaps are each equivalent to one 10pF capacitor in series. In the idle case (not touched), you'll get 10pF (gap) - 10pF (plate-to-plate) - 10pF (gap). That's three 10pF capacitors in series, which equals 3.3pF.

In the touched case, you'll get 10pF (gap) - 100pF (plate-to-plate) - 10pF (gap). These three capacitors in series give you 4.8pF.

Now your 10:1 difference decreased to just 1.45:1, which is even worse than the self-capacitance sensor with the same air gap.

We can also look at it this way: If gaps would increase the sensitivity of capacitive sensors, those sensors would be able to detect someone touching a plate somewhere on the moon with the sense electrodes being down on earth - the bigger the gap, the better the sensitivity, after all. That's quite obviously not the case.