I did another try and compared a self modulated Oscillator (OSC1 -> OSC1 Coarse Frequency) with an Audio Out modulated one.
When using two modulation slots both with "Audio Out" as source, then the modulation is doubled and comes pretty close.
This leads me to the assumption, that "Audio Out" may just be a little quite and needs doubling in some cases.
As soon as filter drive is turned up, it becomes more apparent.