Hi AB (Erik!!),
Thanks for the post, this is interesting.
Inputs terminated should not make a difference as the input is cap blocked to the input stage, preventing any change to the DC bias condition.
Yes, adjust in the connected situation, I agree, that's good sense.
If you reduce R3, from 39K for example to 36K, more current flows into the input stage and offset falls negative, that is, becomes more negative wrt ground.
If you increase R3, offset rises, that is becomes more positive wrt ground.
If you can look at the offset with a CRO, you may notice some AC (HF) on the line, and this might upset a DC measuring instrument.
Otherwise, it indicates to me that there might be a leakage issue in the input cap - which on the N+ is a 22uF BG NX cap, known to be leaky.
BTW, 130mV on a speaker is nothing to worry about. It amounts to 115 mW dissipation (Eexp2/R) and with most modern drivers designed to handle up to 100 watts continuously, this does not cause any problem at all. It does displace the cone very slightly, but it's trivial and does not affect the sound quality at all.
Hope this helps,
Hugh