Let's try Caesium
I've just tested the whole of the original Eternal demo pack with this new version (the "warp directly to the next stats screen" feature made it fairly quick and simple to test nearly two day's worth of demos).
There are three maps where spechits overruns occurred (including et31, as mentioned earlier):
et12: LinesID: 1350, 1355, 1356, 15, 1349, 16, 1348, 1362, 1361.
et25: LinesID: 638, 1910, 1911, 1914, 1915, 1916, 1917, 1919, 1920.
et31: LinesID: 2921, 2922, 2923, 2924, 2933, 2936, 2942, 2943, 2965.
et122614.lmp plays back OK with prboom-plus both when it attempts to emulate the overrun, and when it does not.
The et25 demos (et252031.lmp and beginend.lmp) desync near the start, but they also do so with Doom2.exe (possibly they were recorded with a different version of the wad or exe), so it is difficult to say for sure if it is being emulated correctly in this case. The desyncs look the same though.
Edit: Found another possible overrun in Memento Mori:
mm28: LinesID: 2022, 2023, 2025, 2015, 2016, 2017, 2018, 2019, 2020.
This one seems benign, and of the demos I tested, only occurs in 30mm8356. It doesn't cause a desync there though, whether an attempt is made to emulate it or not.