If you want to kow what resolution  you can achieve, one way in l2-regularization problems is to construct different forward problems with different distribution of mesh points. If the problem is z-direction, fix x and y mesh distributions and change z only. Then, make a singular value analysis study of the resultant matrices. If you plot the number of useful singular values of each matrix against the resolution on z depending on your mesh in each case, you will see there is a moment where the number of useful singular values does not increase more, this is the resolution that you can achieve (usually the z-resolution of your tommographic system). I have a proceeding about this:  "FDOT reconstruction and setting optimization using singular value analysis with automatic thresholding". DOI: 10.1109/NSSMIC.2009.5401647 and a little more written in my thesis, if you think it can help I can send you them.

If you want to improve it, I know there is people using depth compensation algorithms as: Niu, H., Tian, F., Lin, Z. J., & Liu, H. (2010). Development of a compensation algorithm for accurate depth localization in diffuse optical tomography. Optics letters, 35(3), 429-431. 

But I have not tried directly.

Hoping it helps ;)

If you add to your singular value analysis a study of the Picard condition, it will be more accurate since it takes into account  not only the forward matrix but also the data ;)

If you just want a quick way, and just a rough idea: flip your output on its side (e.g. xz view) and do an FFT. There will be a bright circle of information(like when focusing), where this runs out your information runs out. You'll see the missing wedge nicely:) if you did dual tilt then you will have some in the z.

However, if you have a neg stained resin section you would be lucky to get 3nm resolution (you may of course have contrast), especially in Z, which might explain why you can't find it.

Thanks for the feedback. I am doing a negative stain on my sample and it is a struggle, especially when I am trying to get a consistent ~3 nm height. I will try the FFT method and see how it goes.