Peter Johnstone has pointed out to me that there is, of course, a much easier example of a category in which projections from products needn't be epi: the good ol' category of sets. As they say, the empty set really does exist.