We must examine the feedback transmission from the output of OPA to its inverting input - Nyquist criterion. It turns out that the transfer of transistor (from the point of small signal properties) is very interesting - CB connection !!!
While a diode is not about bigger problems. Just specify its signal (small-signal) resistance.
For negative input gradients, the transistor-feedback amp is simply 'slow'(er than the diode).
Positive gradients are interesting: when the gradient exceeds the transistor's transit frequency, nice oscillations occur. (Starting with distortions at lower frequencies.)
The diode LogAmp is stable and the response times are symmetrical.
Josef - interesting problem, indeed. I agree to your conclusions.
I suppose your simulations are based on a unity gain stable 3-pole model for the frequency dependence of the opamp, correct?
(1) In this case, we can replace the diode with a resistive element (representing the dynamic resistance at the diodes operating point). The circuit will be stable because the loop gain will always be somewhat smaller than the opamps open-loop gain (and the closed-loop gain in the critical region will still be above unity). The diode capacitance will not adversely affect the stability properties (highpass effect in the feedback path).
(2) In case of a transistor in CB configuration it is the gain of the transistor stage which matters. And this gain depends on the quiescent current of the transistor which is set by the DC input voltage Ui. Hence, the loop gain can even be larger than the open-loop gain Aol of the opamp alone. This can shift the loop gain to a value of unity for the frequency where the phase shift reaches -180 deg. Therefore, the closed-loop can be unstable for DC input voltages above a certain limit.
More than that, for rising frequencies the transistor stage causes additional negative phase deviations within the feedback path (lowpass effect). This can even make the situation more critical.