Firstly, I must apologize for coming a month late to this party but I only just stumbled across this thread on a Google search for something else.
It sounds like the problem is something I came across and did a bit of research on. The problem is not with piezo sensors, but with the dynamic behavior of the bed being unpredictable in much the manner of the sand plates (Chladni plates) beloved of high school science demonstrations.
I found two ways to combat this:-
The underbed sensors are much more predictable if they are placed beyond the periphery of the probing area - having three closely matched piezos helps a lot as well.
Use a single underbed sensor to get the accurate nozzle contact and only probe directly above this sensor. A second sensor of the contact type such as a BLTouch is used got get other data for bed leveling/mapping. This does require that there is some way to select the sensor to tbe used for each purpose.
I have had a great deal of success with the second method and am able to get results with an accuracy of better than 3 microns at all points on the bed.