There was actually another point I intended to make about IRV, which I forgot as I was rushing to process 103 e-mails accumulated over the weekend.
I think the arguments about nonmonotonicity and complexity can be turned back on the argument about vulnerability to fraud. In the current system (and some others), everybody knows how to stuff a ballot box, and that operation can happen at any level. Given the complexity, and especially the nonmonotonicity, of IRV, no one could be sure what the effect would be of any particular change in a vote. Assuming that fact were generally known, the only possible locus for fraud would be at the final tally, which would have to be done by computer. Watchdogs would thus be needed only at the final step, rather than at individual precincts all over the country. Thus it looks to me as though IRV would be least vulnerable to fraud.