Seems it was a misunderstanding what was measured (clarified in the post):
I measured 7V after buffer and the noise added by the LTC1052 is quite high (DS: 1.5µVpp typ, 7Vdirect -> 7Vbuf: ~0.7µVpp), did not realized that it adds so much noise...
Battery powered 0.1-10Hz noise:
7V buffered: ~1.7µVpp (12s, typ) - same with linear bench psu
7V direct: ~1µVpp (12s, typ)
Seems it is in the ballpark of other users
There was no measurable difference in noise with or without air-draft excluder in 0.1-10Hz range for this unit, but maybe it affects noise below 0.1Hz
TC compensation leads to next project: thermal chamber
Or even two for direct comparison of temperature stabilized one against DUT?