Well. Simpson method gives exact result while integrating cubic polynomials which of course does not depends on n. Try, for example:

(simpson-integral #’cube 0 1 2))

It will give you 0.25 - exact answer.

(simpson-integral #’cube 0 1 100) returns inexact answer because of inexactness of ‘float’ computations - cube function is calculated with limited precision.

(define (wallis-f n)
(/ (* (* n 2)
(* (+ n 1) 2))
(square (+ 1 (* 2 n)))))

(define (square x)
(* x x))

(define (next x)
(+ x 1))

(define (w-pi n)
(product wallis-f 1.0 next n))