SO to continue my previous reply
After this I made some more test with even lower gear as elite949 suggested.
I will spare you all the test results (about 2hrs of testing and comparing) but will show 2 more (final) results. I also played with power smoothing. And those 2 picture also show the effect of this parameter.
First is ERG interval, keep RPM of about 90 power smoothing set to 1, 34x28 gearing:

- ERG 90rpm 34x28 sm1 (Klein).JPG (39.47 KiB) Viewed 10119 times
And after changing the power smoothing back to default (2). Increasing it even more will lead to more noticeable delay when stepping the power

- ERG 90rpm 34x28 sm2 (Klein).JPG (40.49 KiB) Viewed 10119 times
You can clearly see the effect of the power smoothing, it dampens the variation in power read-out
This is actually the best I can get in terms of respons to a power change. And I would actually be quite satisfied with this performance. I mean look at the first 5 steps. Nice. Ok, the steps down are not good, but they do not really matter. I can live with that. But what happened in the last 2 power increments? Here we see a noticeable delay again. I saw this often. I can not help to think that the delta in power step might be of influence as well.
So what did I learn of all my testing:
1: The respons of the trainer can be good. Because when you set the load manually (POWER mode) the respons is very quick. I think the ERG adds another control loop in the trainer, and that seems not to be tuned perfectly.
2: For now, when entering (short) interval sessions, maintain a low gearing(34X28) and try to maintain a high rpm (~90).
3: Elite is not there yet! Performance must be improved. I hope they listen to us.