Official comment

It all comes down to how well the scene was reconstructed in the 3D model. The placement of the GCPs doesn't change the model but really only serves as a place that we use to align it. That is why we do the three anchor points. If you do more than 3 anchor points, the error between the model and the anchor points will be minimized but none are guaranteed to actually line up.
The curvature you describe is an artifact of the camera calibration that we compute being slightly off. More overlap will give us more points to work with and our recent versions of the app have been giving us better images to work with to create the calibration, but it is always possible to be off by a bit.
Basically, we do as well or better than any other software given a set of images, but the elevation numbers given in our outputs over long distances can be off. This is the state of image based the 3D reconstruction. We put the priority on the 2D being accurate at the cost of the 3D.