Let's try to answer the question with science. I created a simple web scraper (yeah, scraping is usually a bad idea when there is API available, but I don't think that's an option here).
Trying to ignore some weirdness in the results (probably caused by caching, or something like that; see below), I found out that there are 4 questions that are in questions/unanswered, but not in unanswered/votes (which exactly matches the difference between 996 and 992):
What these questions have in common is that they have negative vote score. It looks to me like those are the only four unclosed unanswered questions with negative score, which would explain the difference.
The actual results, were, well, weird.
First, among the, say, 992 questions, the number of distinct questions was sometimes 942, sometimes 892 and sometimes even 942. Since it's always a multiple of the page size (50) less, this leads me to believe that the order of pages is not always consistent and that a certain page can appear at at least two different positions.
Second, the total number of questions/unanswered was sometimes 996, but sometimes 992. I don't understand how could this happen. (The number of unanswered/votes was always 992.)