I only donate my time to reviews towards conferences and journals which allow unlimited public electronic dissemination of authors work (including the post-review modifications) without charge, or with minimal charge .A minimal charge would constitute a conference fee of under 400 pounds for conferences (including conference rooms and organisation, but excluding food and lodging), and a publication fee under 50 pounds for journals (which don't have the cost of putting on a conference). With rare exceptions, I will not do reviews (without charge) for a journal which restricts an unpaid author's right to freely disseminate their work unless it provides some form of free electronic access to all. Each review will involve a required donation on the part of the journal of 400 pounds (or 1/4 any stated open access fee, whatever is greater) to a charity of my choice. This covers some of the cost of the time involved in doing the review. Journals which do provide free access or unrestricted right to free dissemination (with the obvious citation/formatting caveats) will have priority for reviews and all reviews will be free of charge. I believe the result of work and reviews paid for by the public purse (or by an agency funding public research) should be fully and freely available to the public. For an example of a journal which successfully operates this policy, see the Journal of Machine Learning Research.
Due to the large number of review requests I receive, I also restrict my journal review numbers to at most five times the number of journal submissions I make, and my conference review numbers to at most eight times the number of submissions I make. I prefer to give a fair amount of time to reviews out of respect for the work that the authors have usually put into the paper. This restriction helps to allow me to do this. If only a cursory review is required then this can be requested (for example if corroboration of another review is required).