I guess your Q1 had drain and source connected backward. Should IRLML6401 be a negative drain device? Though, I can see by a strange way Q1 as is will still work: when no 3.3V supply, the source of Q1 is initially pulled up through the diode in it and that give the needed negative Vgs. Also, a small cap between VBACKUP and GND may ease the supply transition and help to turn on Q1.
As l mentioned you might even replace Q1 with a schottky, or otherwise R1=1k is way too small. It drians the 3.3V supply unnecessarily. Make it 1M should be fine.
Q2 is acting not more than the diode in it. And worse yet, the diode has 1.2V drop(max). A schottky is better. Did you measure the voltages drops over Q2, or did I miss something?
Edited: