[G6Tanuki]

Ferrites can be designed to have positive, negative, or pretty-close-to-zero temperature coefficients - so don't dismiss them out-of-hand. In the past I built various direct-conversion receivers using ferrite toroids for the VFO and the drift was perfectly acceptable for listening to SSB/CW stations on 14MHz [SSB/CW have at least an order-of-magnitude greater sensitivity-to-drift than AM].

For me, the big thing is to keep the transmit VFO running all the time to prevent the thermal-cycling effects which come with stopping the transmit VFO when you're on receive. OK, you say, but won't leakage from the transmit VFO get back into the receiver and cause heterodynes/desense? Yes - but adding a diode switch [semiconductor or something like an EA50 according to your personal orientation] and a capacitor can happily shift the VFO a safe couple of hundred KHz away when you're on receive.

Alternatively, run the VFO at 2x the intended frequency and use a divide-by-two circuit. You can leave it running all the time then, and the divide-by-two will divide the drift by two too.

And doing other things - like mounting the VFO tuned-circuit in a nice solid thermally-stable diecast box - can probably do more to promote stability/long-term-resettability than fretting about coil characteristics. Short-term drift has rarely been an issue for me with VFOs below 5MHz; the more-annoying thing for me is long-term drift which compromises resettability and makes a nonsense of carefully-constructed calibration charts. Cooking a VFO in the oven [yes, heating it to 150C] several times over a period of days can help stress-relieve the construction and promote long-term resettability; I've got one here running on 4.7MHz [EF91 electron-coupled Colpitts, ECF82 cathode-follower/buffer] that settles to within 100Hz within ten minutes from switch-on from cold.