Again, I'm not totally familiar with this topology, but it looks to me that if your LO is in the small signal regime for the lower diff pair, then the bias current to the upper pair won't change. In that case, you just have a differential gain stage for your RF input, which explains the ~10dB of gain you're showing in your second plot.
If you hit the bottom pair with a high power LO, however, then you'll end up pulling pulses of current through the upper pair, with two pulses for each LO cycle (hence the x2 operation). That will then multiply the RF signal, and you'll get your 2xLO - RF signal.