The following animation demonstrates behaviors that may greatly depend on internals of how a program runs, if not the environment.
I reproduced this in a 2019-era map of mine. It's probably not surprising that I didn't observe this earlier, given that it's apparently most visible starting from an early 2022 build of EDuke32.
So, as should be known by now, the Build Engine's original software renderer has limitations when it comes to rendering walls whose tiles don't have power-of-2 heights in pixels, say 11 pixels.
But how may it look like across different renderers, or even just the software renderer with different revisions? I'll concentrate at the software renderer here.
See the given animation and reference map, using tile 799 with the height of 11 pixels (I originally used this tile for a stairway):
1. The first frame (an otherwise black-colored column having a somewhat brighter pixel on the bottom) matches EDuke32 SVN r8071 and earlier revisions, and also matches DOS v1.5.
2. The second one (all pixels being black-colored) matches SVN r8072, along with r9899-3b9f04270. At far as I can tell, it still matches Linux release builds of later revisions of EDuke32.
3. The third one (bright red-colored column) matches Linux debug builds (RELEASE=0 OPTLEVEL=0 in my case) of r9900-629b3f330 and later revisions.
For reference, the impacting commits are:
r8072 (SVN) - "Align cache1d memory blocks to system page size"
r9900-629b3f330 - "mimalloc: update to 6ead2840ec620f1fc06297c310c902e56835c0a7"