as I already said, the traffic models are not properly implemented in the link level uplink simulator. Further, I am not sure which simulation results you are showing me here. You simulate over 15 fixed CQI values for 30 values of SNR? Is this the sum troughput?
I advise you to run a simulation with only two users, no traffic models and the PF scheduler to see if this scheduling algorithm behaves the way you admire.