In design, I am finding that the IIP3 of a true class AB opamp is worser than a pseudo class AB stage for similar power dissipation.
Linearity is quite a topic itself, while your statement is tricky to judge without investigation of your results.
What do you mean by true and pseudo class-AB stage? Is pseudo one, the class A bias structure with AC coupling for signal?
In general, input differential pair has non-linear transconductance, current mirror has limited linear range and conversion to output voltage is done on nonlinear current source. What makes OPAMP linear, is a negative feedback configuration, suppressing input signal swing by gain of amplifier, so changes of operating points of each device inside opamp are so small that we considering their higher-order terms as negligible.
And now is the first point. If we consider class-AB biasing (lets look on Monticelli's as is very common), for quiescent conditions opamp achieves minimum gain (first stage load is a sources of floating current source structure, so OL gain is like single stage only), which is boosted at high output load case (floating current source become high impedance for signal and gain can be doubled in dB).
So, if we suppose that the main source of nonlinearity is input differential pair, class-AB opamp has an extra nonlinear term related to gain variability with signal.
However, it can be different story for different class-AB solutions.