Comment by tromp

If it were about coding fast growing functions, then it would have had to mention the incredible 47-bit lambda calculus term λn. n n (λe λx. x e x) (λm. m (λe. m e m)) that achieves f_ε₀ growth. But it expects its argument to be a so-called state numeral n, rather than a Church numeral, and even state numeral 2, which is \e\f\x.f e (\e.f e x) is already 32 bits long, making the application take 2+47+32=81 bits, more than the required 64.