I don't read that from the paper - he shows that 45 degrees is better than 35 degrees, but doesn't really explore different speeds, or any other bank angles - so it hasn't got as far as a general case.
Oh, I think he does. At the end of the section "the optimum bank angle" he shows that the rate of height loss per unit angle turned (in the steady state case) is proportional to 1/sin(2*bank), which is minimised at 45 degrees. The only reason he shows 35 degrees is because it's Eckalbar's recommendation.
Of course one could argue that the true optimum comes from a non-steady state turn, but I don't think you'd do much better.