Torchaudio; resampling; Performance Benchmarking

I’ve read the tutorial here, and tried to understand the sentence :
“using kaiser_window results in longer computation times than the default sinc_interpolation because it is more complex to compute the intermediate window values - a large GCD between the sample and resample rate will result in a simplification that allows for a smaller kernel and faster kernel computation.”

The results down there say that
“16 → 8 kHz” is slower than “48 → 44.1 kHz”. But I guess a GCD of 16,8 is 8 and that of 48, 44.1 is around 4 so GCD of 16,8 is larger and it should be faster according to the tutorial. Can anyone explain this?