This paper shows that, in certain high dimensional integration problems, quasi-Monte Carlo using Sobol sequences can never be more efficient than the ordinary Monte Carlo no matter how carefully the primitive polynomials and initialization numbers are chosen