I found my programs and boy, my memory's apparently not all that accurate.
It turns out that:
1) The conversion was rather more lossy than I remembered. The 3 terrain changes you identified, I "corrected" in the conversion program as explained earlier, but the rest I apparently did not have any build-in corrections in the program, so either I ignored the differences because it doesn't affect level solutions (ie. terrain is terrain even if the pixel's color is wrong), or I did apply manual corrections afterwards on a case-by-case basis for some of the other differences. It also looks like some interactive objects in a level may occasionally have been dropped or otherwise modified intentionally, due to reasons such as because DOS may glitch or crash on objects whose trigger areas fall partly or wholly outside the level area boundaries. (Perhaps that explains Sunsoft 11?)
2) Apparently for some levels, I intentionally had the conversion program deviated slightly from the norm when calculating positions. Unfortunately I did not comment on the exact reasons in the source code, but I'm guessing it was probably somehow necessary or highly preferable for playing the level in DOS Lemmings. So they are apparently not mistakes per se but "by design", though definitely not valid reasons for your case. [In case anyone's wondering: the levels with adjusted positioning during conversion are Tricky 12, Mayhem 1, Present 10, Sunsoft 5, Sunsoft 20.]
Anyhow, according to the source code, the following terrain are different between DOS and Genesis:
set 0: 36 42 43 45 46 47
set 2: 40
set 3: 32