Short answer: One day in a month.
In a piece on the "challenges and opportunities" of peer-review, Ware (2011, 25) estimates the median time spent per article to be 5 hours, and the average to be 8.5 hours. He acknowledges that these figures "vary by discipline and also by experience" (ibid.), but doesn't go into more detail. "Active reviewers" report an average of 14 reviews per year (ibid., 28). There is no explanation in the paper of what counts as "active" or how many papers the average academic reviews in a year.
However, in a follow-up publication, the same author defined active reviewers as those who review more than one paper per month (Ware 2016, 3). The same publication reports a follow-up survey with a slightly different methodology. Although the findings are therefore not directly comparable, their thrust is similar, with the average (mode) number of papers reviewed per month between one and two (ibid., 33), the median time spent on the last review at five, and the mean at 8.4 hours in 2017 (ibid., 34). The survey included more than 2000 respondents from different fields and demographics, with the arts and humanities somewhat underrepresented (ibid., 45).
Figures: Ware 2016. CC-BY-NC-ND.
Ware, M. (2011). Peer Review: Recent Experience and Future Directions. New Review of Information Networking, 16(1), 23–53. https://doi.org/10.1080/13614576.2011.566812
Ware, M. (2016). Publishing Research Consortium Peer Review Survey 2015. Mark Ware Consulting Ltd., Bristol, UK. http://publishingresearchconsortium.com/index.php/prc-documents/prc-research-projects/57-prc-peer-review-survey-2015/file