fixing this is rather easy, for example using 8000 cycles instead of 7200 would certainly do the trick. But if that's what you did, the questions are : will it work in all cases now, and could it create regression ?
I'm afraid it could be the case, that's why unless I find time to really measure various ikbd commands on STF, I prefer leaving it this way under Hatari, changing a value based on a single demo doesn't seem a long term solution to me (in my experience, it often created problems elsewhere).
So, I prefer a demo that is known not to work, than trying to fix it by changing values that were not really measured, else in the end, the source code for Hatari will become a giant piece of workarounds and empirical constants with no real meaning.
But of course, you're free to modify Steem the way you want