I think it could be done in multiple ways. I'm sure you should be able to find some configuration examples out there where people have set up Linuxcnc to control knee mills with both the quill and knee contributing to the Z axis position, this would be nearly identical.
It might be better to simply do it using X and U axis, where U would be the extra stage mounted on the X. This set up would give you better control over when and how each axis is used. Alternatively you could simply use a Kinematics model were X is the sum of the two X axis joints, but this might give you less control over when or how each joint contributed to the position and movement of X.
It may even be possible to configure it using a custom Kinematics where the X-axis = joint0 + joint3 and the U-axis = joint3. Allowing you to use U commands to control the extra stage's position, but still have it automatically contribute to the final X position.
(I have never actually tried any of this myself.)
All that said, is this really a good idea? Adding a lot of cantilevered forces and movement beyond the machine's designed structure could have less than desirable results (such as magnifying/multiplying tolerance errors and greatly reduced rigidity.)