Point taken about outer device and its variable heating. I had noted that the drain voltage came from amplified reference in the same way that many linear regulators power their reference from their output.
Yes, I'm aware that the sensor needs low thermal mass and good coupling so that it accurately reflects what is going on. Also about size and power. Ultimately, there's a thermal mass whose temperature you want to change by heating, so the lower the mass, the less power needed. At the same time, there are losses to the environment, and although it would seem sensible to minimise them, they are actually needed to help in stabilising temperature, because if you apply too much heat, you want the system to cool quickly (ideally with the same time constant as for heating). Subconscious wondered about mechanical stress to the JFETs and gluing in block. I'll think about that one some more.
Later edit: Conscious has thought about it and decided that gluing JFETs into a block held at constant temperature is fine. What wouldn't be fine would be gluing them into a block that changes its temperature and has a different coefficient of expansion to the epoxy containing the silicon die, thereby applying variable stress to the silicon as temperature changes.