The updated experiment only modifies a state variable (GKr), not a model parameter.
the im_param option is intended to modify model parameters, not state. Model parameters are (mostly constant) variables influencing how a model works. On the other hand, state variables change in time and contain a current state of a model. Some models have variables being parameters and state at the same time.
It is important to keep in mind that im_param modifies a model, as if the model specification would have been changed, therefore, before it is applied to the concrete experiment. As such, your experiment setup first redefines the starting value of GKr in the TT2 model, then chooses to restore a previously stored state, thus overwriting the state variable modification.
If you change GKr for GNa, which is a pure model parameter, you will see that the results are indeed different when restarting.
This is a very subtle thing. I will add an issue, to parse the user input for this setting and issue a warning.