Just to be clear:

OP has one computer
OP has four bikes, and each has a mount for the computer
Three of the bikes work perfectly
One bike works poorly
The poorly working bike has a different model of mount
The OP has a different mount that will allow the computer to work perfectly on the problem bike
All of this seems to point to a problem with that particular mount

The computer elevation failure is a very specific, repeatable problem
The computer does record altitude in the correct direction, but at a much high rate than it should
In addition to elevation, the computer also has incorrect, wildly swinging grade outputs

I am not familiar with this computer, but I'll take a guess that fits the above observations. I image the computer has three ways of determining elevation - GPS computation, relative barometric pressure, and dead reckoning based on speed and grade. It likely monitors these three things continuously while it's moving and outputs to the user what it thinks is the best information without bothering to tell the user what that output is based on.

I think the elevation is a symptom. The real problem is the angle of attack sensor in the unit. The sensor works fine in most mountings, but for some reason acts erratically in this particular mount - perhaps it just shakes or resonates the wrong way. Nothing in the computer is broken.

The computer unfortunately is not smart enough to realize that the input from the angle of attack sensor is garbage, so it weighs that input heavily in its elevation calculation based on whatever algorithm it uses.

The rest is obvious. The bike shakes the mount. The mount shakes the computer. The computer shakes the angle of attack sensor (it shows this on the screen). The computer uses this overly rapidly changing angle of attack data to dead reckon an elevation calculation. That dead reckoning is 4x too high because the computer thinks you are constantly going up and down little tiny hills.