Problems with simulation of a common source with current mirror load

I'm trying to perform an ac analysis on the circuit with LTSpice and level 1 parameters. From the theory I should get a pole at frequency 1/(RL*(Cgd+Cl)) and a zero in gm1/Cgd. The gain should be gm1*RL, where RL=[(lambda1*I)^(-1)] || [(lambda2*I)^(-1)].

(W/L)1=20/1; I=77.5uA; kn=110uA/V^2; lambda1=0.04V^-1; lambda2=0.05V^-1; RL=143kOhm; Cl=100fF and then Cl=27pF

Now, the questions are:

1) is Cgd=CGDO * W with W=Weff?

2) as I have CGDO=220pF/m and a transistor with W=20um I should have Cgd=4.4 fF but from computer simulation Cgd=26.5fF (as I get the zero at 22GHz and gm1 is 584uS). I don't understand why... is the Cgd of 4.4fF wrong? Is the Weff so much different from W?

3) If I use Cgd=26.5fF and I apply the Miller effect, the pole I calculate with Cl=100fF is close to the one I get from computer simulation. If I try to increase the Cl up to 27pF this is not true. In fact I need to increase the value of Cgd to get a pole frequency closer to the one of LTSpice. But this is wrong, because the zero stays always at 22GHz and Cgd doesn't change (as well as gm1)

So, generally speaking, I do not know how to explain the behavior of the simulation and I can't "chek" the hand calculations given by the theory.

How can I solve the problem?

SC con specchio analisi in frequenza.asc