Because the output has an upper limit of slew rate (SR=dV/dt).
For sinusoidal signals, the highest slope occurs near zero which is proportional to Vp and the frequency.
V=Vp*sin(w*t)
dV/dt = Vp*w*cos(w*t)
where w= 2*pi*F
And
dV/dt max = ± Vp*2*pi*F
So for the output to give the expected Vp the following inequality should be satisfied:
Vp * 2 * pi * F < SR
F < SR / (Vp * 2 * pi)
So the lower Vp we accept at the output (by decreasing Vin), the higher the frequency we can get.
Edited:
If we let Vp_in constant and we increase the frequency, Vp, at F = SR / (Vp * 2 * pi) where Vp=Vp_in*Gv, will start to decrease as if it is Gv decreasing. The new Vp tries to satisfy:
Vp = SR_max/(2*pi*F)
Recommendation: Try to become more familiar with data sheet information.
The given voltage values (pp) are not maximum values but measurement conditions- as indicated at the top of the relevant columns in the data sheet.
Nevertheless, the answer from KerimF regarding slew rate influence is valid and is the reason for not driving the opamp at its thresholds during these measurements.