I am designing a double balanced Gilbert cell with current bleeding to increase the conversion gain and to reduce flicker noise, to use smaller switches .

My problem is this:

Every one says that linearity of a Gilbert cell is primarily determined by the linearity of the transconductor.
In my design, the transconductor alone has IIP3 = -10 dBm.
Overdrive of the switches = 50 mV, LO drive = 0.2 V peak single ended.
I_bleeding = 0.5 * I_transconductor.
But my mixer IIP3 = -18 dBm.
If the switches switch properly, they should not affect the linearity.
What could be the reason for this IIP3 degradation?