These questions are typically not asking for percentiles among students in any particular course. Rather they are asking for comparisons among all students you have interacted with. Of course this is also vague and contingent -- in theory, someone who has only interacted with students at absolutely top programs should give lower ratings than someone who had interacted with those same students and also a wider range of students.
Thus I agree that the numerical rating system is too flawed to be taken very seriously. I think it just gets the idea across that recommenders should be thinking about how the recommended student compares to other students at a similar point of their careers and including comparative information.
Giving a student who got an A in your course a "top 25% rating" because their numerical performance is in the top 25% of your course seems to be both an overly literal interpretation of the percentiles and also thinking too narrowly to be helpful. Again, you are not being asked to report on the student's performance in just one course, and if you really know nothing about the student other than the one course they took with you than you are not an especially good person to write a letter for them. How did the student do in other courses at your university? More importantly, how good is a student who gets an A in your course but is not at the very top of the course? Was this student's mastery of the material not as strong as the top students in the course in a way that you think is meaningful in the context of graduate work, or is the situation that you have many excellent students and someone has to come out on top percentage-wise? You also mention that you teach at a top 10 university "if it matters." Yes, it matters! Much as you suggest, it is probably the case that an A student at a top 10 university who is not at the very top of the course compares well with all but the most outstanding students at a lesser university. But it is your job to figure out to what extent that is true and argue for it in the letter.
If you are asking what is usually done on these numerical rankings: most of the letters I receive at my top 50 US math PhD program rank students in the top 10% in most of the categories. Ranking a student in the top 25% across the board will come off as unsupportive unless you explain yourself very carefully in the letter. I think you and the student you recommend will have an easier, better time if you choose among the first three categories in the numerical ratings (the way it works at my program is that students are rated separately on a handful of categories).