Nobody seems to have studied this and there is no proof either way. A full-time, rear-facing noise maker is probably not the most effective conspicuity measure, and it does have community noise issues, but nobody can say that they are not of some help in getting attention and reducing crashes. Hurt was inconclusive on this issue, as there were slightly more (30.1% versus 27.3%) modified pipes in the crash sample. The difference is probably not significant and the types of exhaust system modifications were not broken out.
Maybe the current OSU Motorcycle Crash Study might actually prove that loud pipes do save lives (or not) and settle the matter? At the reduced sample size of 300, this might be the only fact that gets proven in the compromised study.