There is indeed an upside, though in my opinion, it does not justify the amputation of healthy, functional tissue in infants who clearly cannot consent to it and condoms are readily available even for these with allergies to natural latex rubber.

The most recent studies that I've read did elucidate a likely mechanism too. Making the glans an external organ, rather than be protected by the foreskin, causes the development of keratinous tissue (literally called "horny" tissue) on the glans in order to protect it from the environment, rubbing against clothing, etc. Effectively, it becomes callused. The horny layers are composed of dead and denucleated cells, creating a physical barrier that bacteria and viruses must pass in order to infect the underlying cells.

Note, though, that there were three studies conducted in Africa on the impact of male circumcision that was/is cited on HIV prevention that are so blatantly terrible tha PLoS Med and the Lancet, along with whatever IRB was in charge ought to see reparitive and punitive fines brought against them. The studies show extraordinarily poor study design, data collection, data analysis, and alarming degrees of multiple biases. The issues include, among others:

All HIV infections were assumed to be sexually transmitted and the result of heterosexual intercourse (bizarre assumptions). Conservative estimates from follow-up research puts the percentage at only 43.1% of the infection from all three studies being sexual transmission, with no extant data or tracking on partners involved. Due to not accounting for the vector of infection, it is impossible to draw the causative relationship that the researchers claim.
Improper controls: The test group were given sexual education around STI transmission and proper condom use. The control group were not.
Lead-time bias: Data collection began immediately, despite researchers instructing the study group not to have intercourse for 6-8 weeks and likely discomfort with intercourse and increased condom use occuring in some who undergo adult male circumcision up to 12 weeks following the procedure.
Attrition bias: Significantly more subjects dropped out of the studies than became infected, which was not accounted for appropriately, corrupting the dataset used for analysis.
Duration bias: The PLoS Med study was planned to take 21 months of data but only ran for 14 months. The Lancet studies (near identical to each other) lasted 24 months. Neither is sufficient to either remove tye statistical significance of the lead-time bias, nor to provide objective long-term efficacy rates for an irreversible treatment.
Expectation bias: A number of principal investigators involved in the studies had previously publicly called for mass circumcision campaigns. This alone is a major red flag that should have resulted in more critical review of the study protocols and required that they, at the very least, mak, clear disclosures of their personal biases but, to have actually trustworthy results, they should have had no role in data analysis due to clear lack of objectivity.

Referenced studies:

PLoS Med 2: e298. doi: 10.1371/journal.pmed.0020298
Lancet 2007;369:657–66
Lancet 2007; 369:643–56