I had a bit of a further play with it today and logged the value of the LM335 with my K2100 every second. The plot & data is attached.
Took about 8.5 minutes to reach -6°, took 3.3 minutes to go to 38.5C from there. This is with 10V across an element capable of 15V.
I don't like the drifting once it closes in on to the setpoint, will have to investigate further. Logging and plotting was done with python.
Edit: I did a second run with higher P-Gain to see if that would help, it looks like that to me. Setpoints were a bit different which hinders comperability. I also added the complete plot, but there is not much to be seen there i think.
Edit2: Upon further pondering, i believe part of the problem is that the part of the voltage coming from the LM335 i try to regulate upon here is already in the 100's and 10's of uV. Also, for a remote sensor, a current based one should be better with regard to signal integrity. I think i should retire the LM335 together with the current mirror for a LM334 and an OP07 in transimpedance configuration.