Hey there, it's me again!
I noticed at least two mayor problems in the SPICE model of the ADA4899. First, the voltage noise is way too low, 0.3nV/rtHz after the 1/f-noise has decreased to the white noise floor.
It's not a problem of Micro-Cap, I already double-checked with Multisim. I could solve this problem by introducing another resistor model with an absolute Temperature of T_ABS=27 and use this model for both R_REs and R_RCs instead of RCOLD! I dont' know if it's a proper solution and I have to admit that I'm absolutely not into spice model design but at least I can now continue doing simulations instead of guessing!
The other fault is input current noise. By adding a 1k resistor before the voltage follower, the noise shots up to 8.1uV/rtHz!!! This translates to an input current noise value of ~8nA/rtHz which is just ridiculous!
By twiddling some values one could also adjust this value, but I think it shouldn't be up to the customer to make the spice model(s) work properly (especially not if there was an external company involved in the modelling process who claims to be a specialist)