I would have thought the situation a good deal simpler than indicated above.
The effect is apparent on all the ADFs I've used from the Bendix as fitted to the DC-4 to the 747-400. Re the comment on towel rail antennae, they are the sense element and don't affect the outcome.
The ADF loop is intended to track only the direction of the beacon signal in respect to the horizontal plane. (no pun) If you've used a wire-wound rotating-loop aerial in sky wave you'll know why Bellini-Tosi, Marconi-Adcock or their successor fixed loops have been in vogue since the days of MF ranges.
While later ADFs perform better (one would hope so) their horizontal reception is still relative to the airframe. I am unaware of an ADF with attitude input, but stand to be corrected. Therefore if the aircraft banks, the ADF receiver tilts.
If you were overhead a station with 90 degrees of bank, the ADF would see the beacon straight down, at right angles, and the needle therefore at 090 or 270. The effect is less, but still there at varying degrees (pardon, no pun) dependent on height, distance and angle of bank.
The effect is regularly seen when turning inbound in a holding pattern, and a pilot who has been watching what he thinks is a good turn onto inbound track, finds the needle does a dart to indicate off track as he/she rolls the wings level.
Trying to anticipate is possible, but due to the variables, it is not easy. The solution is to fly (a) in LNAV, or (b) the turns and outbound leg on DR as accurately as you can. Roll out on to inbound heading promptly, let the ADF settle and correct immediately for track error.