The choice of numerical method in fractional calculus is a trade-off between physical fidelity (long memory), computational cost (dense vs. compressed history), and regularity of the solution (smooth vs. singular at $t=0$). For many problems, the short-memory principle or sum-of-exponentials acceleration is not a luxury—it is a necessity.
$$ a^GLD^\alpha t f(t_n) \approx h^-\alpha \sum_j=0^n \omega_j^(\alpha) f(t_n-j)$$ The choice of numerical method in fractional calculus
$$ 0^CD^\alpha t f(t_n) \approx \frach^-\alpha\Gamma(2-\alpha) \sum_j=0^n-1 b_j \left[ f(t_n-j) - f(t_n-j-1) \right]$$ computational cost (dense vs. compressed history)