This turns out to be more complicated than I thought - as usual.
Yes, something funky is going on. I tried out the last version that I pasted in the previous post and it doesn't work the way intended. It apparently wants to make a dip shape there no matter what. And now it makes a lean problem instead. No knock though. But another problem that becomes apparent is the influence on load:
http://datazap.me/u/tor/tor-204?log=...zoom=2636-2742
Compared to the normal OL fueling (that flat spot between 3600 and 3900 rpm).
http://datazap.me/u/tor/tor-201-1-ia...zoom=5894-5985
It looks like there is some benefit to correct the lean spot above 4000 rpm. This was the more moderate first attempt:
http://datazap.me/u/tor/tor-202?log=...7&zoom=627-725
So as usual, it's probably better to keep it smooth. I am going to revert this and forget about the (perhaps fake) rich dip below 4k and only concentrate on smoothening the lean spot above 4k, which seems to give better load values.
Maybe this dip shape is more to do with load limits?