Although varying the camera distance and focal length will change the object's visual size, it has the drawback of also changing the perspecitve, thus leading to a somewhat distorted view. I suggest to use the camera's SceneScale property instead.
Alas, I have no valid steps to calculate the correct value for that. In my case I have to scale to a cube with varying size, while the window size of the viewer is constant. So I placed two dummycubes at the position of the target cube, each sized to fit either the width or the height of the viewer with appropriate values for SceneScale, camera distance and FoclaLength. During runtime I calculate the new SceneScale by the ratio of the target cube size in respect to the dummycube sizes. This works quite well in my case.
Edit: Here is some code I make for the calculations. ZoomRefX and ZoomRefY are those DummyCubes, TargetDimX and TargetDimY give the size of the current object, DesignWidth and DesignHeight are the size of MyGLView at design time, DesignSceneScale is the camera's SceneScale at design time.
ScaleX := (ZoomRefX.CubeSize*MyGLView.Width)/(DesignWidth*TargetDimX);
ScaleY := (ZoomRefY.CubeSize*MyGLView.Height)/(DesignHeight*TargetDimY);
NewSceneScale := Min(ScaleX, ScaleY)*DesignSceneScale;
The DummyCubes ZoomRefX and ZoomRefY are sized so that they have a small margin to either the left-right or top-bottom edges of the viewing window. The are both positioned so that the front faces match. Also the target object is positioned to match its front face with those of these DummyCubes.
The formulas above allow the window size to be different from design time, but I actually didn't test this feature.