Understand why g1_on_curve from paper is slower than cofactor multiplying

[paulmillr]

https://eprint.iacr.org/2019/814.pdf

The code is slower than this.multiplyUnsafe(CURVE.hEff).isZero(). Need to understand why the "optimization" does not optimize. I mean, hEff aka 0xd201000000010001n has low hamming weight et al; need to understand deeply why the opt is possible.

  // σ(P)
  private sigma(): PointG1 {
    const BETA = 0x1a0111ea397fe699ec02408663d4de85aa0d857d89759ad4897d29650fb85f9b409427eb4f49fffd8bfd00000000aaacn;
    const [x, y] = this.toAffine();
    return new PointG1(x.multiply(BETA), y);
  }
  private isTorsionFree(): boolean {
    // (x²−1)/3
    const c1 = 76329603384216526021617858986798044501n;
    const P = this;
    let sP = P.sigma();
    let Q = sP.double();
    sP = sP.sigma();
    Q = Q.subtract(P).subtract(sP).multiplyUnsafe(c1).subtract(sP);
    return Q.isZero();
  }