I second Lars' suggestion to try it in ScummVM... but might not ScummVM reject it because the resource package hash is different?
Can you describe the changes you've made? Did it work properly without the changes? (Trying to rule out a SCI Companion decompiler/compiler bug here).
Also, make sure you're not loading previous save games (e.g. to quickly get back to the underground) after you've made changes. That could certainly cause corruption.
I've tested it in both ScummVM and DOSBox, and it produces pretty much the same behaviour. I've already made all the changes I need to the mouth animations in the scripts for both rooms (slowing them down, cutting out some of the dead air and keeping them on screen for longer) and can freely test/observe them, provided I do a low item run (picking up as few items as possible). Even then, there's always the possibility that the mouth animations will choke at some point near the end of either of the death sequences.
Going into the underground on a full item run (collecting every single possible item, give or take 1 or 2 items) seems to choke the animations for both the special death sequences in rooms 32 and 33 (possibly related to the current size of the inventory window?) as well as the animation loop I added to the Quit and Restart screens (this I shamelessly pinched from disk version of SQ4).
For the latter, altering the script in Main.sc from:
(method (restart/exit)
(babbleIcon view: 946 loop: 4 cycleSpeed: (* (+ global87 1) 4))
to:
(method (restart/exit)
(babbleIcon view: 946 loop: 4 cycleSpeed: 12)
seems to provide greater overall stability animation-wise when on a low item run, but it still doesn't completely guarantee that the mouth animations for the death sequences won't choke at some point near the end.
I'm sorry I can't be much clearer on this, as it all seems very convoluted and sensitive. I'm just as baffled as you guys, honestly.
