"Shall I poison a well to see how many drink out of it?"
Frankly, the idea of debunking review processes is not that new, and the latest wave started with the Sokal hoax. There were a couple of follow-ups, with computer-generated papers.
The result was a scandal, a journalistic anecdote. It created attention. Been there, done that. It's no longer an original idea. We got it. You can fool badly screened journals and predatory conferences.
But ok, now, you talk about actual "research" with fake paper dissemination. That may or may not be a legitimate research question, but the fact that you talk about "poisoning" arXiv, which is a serious and expressly not peer-reviewed site makes it already very questionable what you are trying to achieve.
It is one thing to embarrass badly screened journals or conferences (possibly even suspected predatory ones - namely the conferences in question) which Sokal and the SciGen people did, but quite another to abuse a perfectly legitimate repository for non-peer-reviewed papers. In other words, it relies on good will and honesty of the submitters trying to submit good quality work. There are minor safeguards in place to avoid obvious scientific spam, but not against antagonistic submitters.
I do not know (and I do not want to know) what you are trying to achieve with this experiment, but the damage that you are likely to do with this can be so substantial in terms of loss of trust of arXiv and preprint services etc, that you would seriously need to pass it through an ethics committee who can evaluate what legitimate experimental parameters for such a large scale social experiment would be.
Without such a permission and clear delineation of your duties and limitations (which your institution's ethical committee has to evaluate), this research will be unpublishable in a serious journal. And frankly, your academic reputation will be finished. Damaging one of the most valuable resources of openly accessibly publications will make you persona non grata in academia.
Let me add one more thing. There are other preprint servers which have a less strong reputation. There, your experiment may do less damage, though also be less expressive. However, even there, I should believe that they rely on good will and that wilful tainting of wells for scientific knowledge is simply maximally antagonistic to the endeavour of science.
TL;DR: I believe the experiment is highly unethical and extremely damaging. However, you may want to see whether your institution's ethics committee can identify conditions under which such a social experiment might be feasible.