why cross section of process p p > z b , z > mu+ mu- is less

If you look at the paper, you see good agreement with the measurement both at LO and NLO compare to MG5aMC.
They also describe how they setup their generation which is much more than a single LO computation since they do a MLM matching with multiple flavor. Looks like they do computation in 4 flavor (in some setup they also use 5 flavor).

I would suggest to reproduce the same setup as the one of the paper and then your LO should actually be higher than the measurement. In any case do not forget that LO cross-section have huge theoretical uncertainty so being above or lower the experimental data is typically not a problem (you need higher theoretical accuracy to start to trust the total cross-section)

Cheers,

Olivier