It still seem to be the current record-holder for temperature stability
Priel Z 1978 J. Phys. E: Sci. Instrum. 11 27?30
got +/-3.5 microKelvin in a well-stirred water bath, which I calculated was the limit imposed by Johnson noise in his - thermistor - sensor.
Sloman A W 1978 J. Phys. E: Sci. Instrum. 11 967?8
He'd published that he thought that he'd run into the limit imposed by the imperfection in his stirring, which meant that he hadn't read the literatur e he cited carefully enough.
Larsen N T 1968 Rev. Sci. Instrum. 39 1?12
only got to +/-15uK but his paper was a classic US National Bureau of Stand ards exercise, with everything done right (and including a ratio-transforme r based bridge and a platinum resistor sensor, which is less sensitive, if more stable, than a thermistor).
Stirring the water-bath does inject heat, and Priel used the stirrer as the controllable heat source in his set-up.
The heat injected varies as the fourth power of stirrer rotation rate, so t he rate of rotation didn't have to vary much.