We know that increasing NA gives better resolution.
A large NA means using a wide cone of view from the lens, with the apex at the point of focus.
We also know that when photographing a 3D structure, out of focus material obstructing the cone of view causes blurred “ghosting”. (The "Transparent foreground" is related, but it's a different issue).
The intervening structure also reduces the working NA.
If looking into a well, half the cone of view is obstructed at the edges of the well.

If the NA is smaller, then more of the floor of the well is seen without obstruction, but the resolution is lower.
The obstructing shoulder marked red, is likely to be well lit. Its out of focus blob covers a lot of the sensor. All the frames in a stack with detail at the edge of the bottom of the well, are severely washed-out by unwanted unfocused light from the shoulder, so very little will show.
]
This was apparent on a recent subject – the peristome of a common-or-garden moss.
The peristome is the developed structure around the opening at the end of the capsule,

when it opens so spores can come out.
A general part-stack view of a peristome

shows its inner and outer teeth, and some escaping spores. Crop, approx 0.5mm fov, 20x objective.
The depth distance between the outer(A) and inner(B) peristome teeth is about 50 microns, measured at the broken tip, (C).
The red square is 50 microns across, just there for scale. The C-A section sketch shows illumination cones for NA 0.42, and stopped down to half diameter.
Squares are 10µm.

A stack was run through the part in question, with an extra exposure after every 4th, at two stops closed, with the flash turned up to compensate.
So on for 10 sets, ie 50 frames which covered the detail I was looking at.
The whole stack was aligned, then partial stacks made from only the full NA shots, then only the reduced NA shots.


Both have USM at 100%, @10 pixels, applied.
The detail on the outer teeth is, as expected, better in the full-NA stack, but on the inner teeth there’s more to see, and certainly less of the objectionable fuzzy voids, in the stopped-down stack.
Animation here , 1.27MB
It’s not too difficult to retouch onto the Normal one, where it's better. Might be better in Photoshop as each layer could then be given a different USM.
An outcome which I’d previously noticed, is that if you want a picture for web use at say 1000 pixels width, you’re better to stop the objective down, because it’ll look better. You don’t see all the fine detail of a higher NA objective anyway, but the high NA ghosting artifacts would still be visible. Same would apply to stereo/rocking images on normal screens.
The workflow is a pain. Exposing a stack with altered lighting and aperture every fifth shot is grim – it’s easy to miscount.
I found that a stack made simply from all frames, didn’t work. I have to use X-Y alignment in the stacker, because my setup is out of line or wobbles too much at 20x. This means that the stopped-down shots were not aligned sufficiently well with the full NA ones, to avoid double imaging.
If the stacks are run separately, then the alignment is further out and it’s harder to retouch, though areas can be stretched in Photoshop.
Ideally, we want something to arrange illumination of selected parts of the subject. A confocal microscope excludes stray light with a pinhole at the sensor. We can't do that but we could reduce the light coming from unfocused parts by not illuminating them.
This could be done above (“Epi”), or subject-side of the objective. But they’re both a little fiddly so I’ll leave them until after lunch.