Josef - I have to investigate the circuit in detail (simulation). My first answer (preliminary): Something must be wrong - as long as you are within the linear region of the opamp, the transfer ratio must be constant.
From Josef`s diagrams I deduce that he was speaking about the performance in the frequency domain only. Of course, in the time domain, we have to face some other limiting effects like supply rails and slew rate limitations.
However, his diagrams clearly show the transfer function in the frequency domain. Hence, the following answer applies to the small-signal behaviour only:
* Josef, I cannot confirm your results. As expected, the simulated transfer function (active L realization) does not depend on the input signal amplitude.
* However, I have observed a strong dependence of the form of the transfer function (active L simulation) from the non-ideal opamp properties.
It is somehow difficult to 'model' an amplitude-dependent L in the frequency domain. Though not completely impossible.
Obviously Josef's problem is created in the time domain, the plots being merely a means to 'show' what's happening. (Step or Dirac responses are quite hard to read. And such diagrams might have given him away :) )
OK - so we have to ask Josef if the shown diagrams result from an ac analysis (frequency domain). If not, the result is no surprise because we always have to face non-linearities.
OZ2 must be able work with U2 - see below. It must not limit. From this we can (and a supply voltage) to determine the allowable UImax. See also https://www.researchgate.net/publication/281625533_Maximalni_vstupni_napeti_pro_zadrze_RC
If it is not valid (see below) the synthetic inductor is "not valid also" - the circuit is non linear - its derivation is not valid. And I would be happy if I could solve the impulse response of this structures with non-ideal amplifiers :-))
Generally, we must examine the level of the signal at each node of the electronic circuit - not just output. In the described circuit it is particularly "tricky" - OZ2 output is not "in the direct path" of signal.
Here is example of LTspice calculations for Notch Filter with Gyrator. The model results for Small signal and Large signal are similare to those at the top of topic.