You specify that your technique would not stand up to forensic examination- why do you assume that if a voting machine election was contested, the FBI, FEC, or state police forensic computer lab would not get involved? Especially if the wrong party got elected.
I think it highly unlikely that the Princeton magicware could do all it is spec'd to do and remain undetectable to a forensic investigation.
Now, how does the magicware know it is being run for the last time?
I offered two methods. The first technique is easier to understand, and gives the general principle, but would probably be uncovered with forensic examination. The second technique (compress part of the original code so that the altered code takes up the same amount of space as the original) would be more difficult, if not for all practical purposes impossible, to uncover via forensic means.
In particular, if someone used the latter approach, all sectors of data on the storage media would contain the same data as they would without the hack, except for those parts which the hacker explicitly changed (such as the vote counts). The sectors may not get written in the same order as they would in a legitimate election scenario, so if the inner workings of the media allow one to ascertain the order in which data were written, it may be possible to tell that something fishy was going on. Even that sort of analysis, however, could be complicated if someone knew how the storage media worked.
Insider knowledge would probably be required to produce phony software that could withstand the tougher levels of forensic analysis. On the other hand, good election systems should be immune to even insider attacks provided there is at least one honest person monitoring them.