I thought I'd have another look at my 20m MFJ Cub this afternoon to see if I could reduce the drift a bit more. I thought I'd try to find out which component was causing the temperature sensitivity by blowing warm air at it through a tube made of a rolled up bit of paper. I could make the VFO quickly drift off frequency by blowing at Q2, the mixer/oscillator chip. But I could do exactly the same by directing the air at the slug tuned inductor. In fact, as far as I could determine, any temperature increase anywhere made the VFO drift off frequency. This reminded me of the time I attempted to use the Cub out of doors. The frequency drifted up and down as the sun went in and out of the clouds that are a permanent feature of an English summer day!
W1HUE's article suggests that it is possible to temperature compensate the Cub by choosing the right combination of different types of capacitor. But it seems to be a lot harder to buy different types of capacitor here in the UK than it is in the USA. You have to pretty much take what you can get, and all of the smaller value capacitors available from the suppliers that will sell to hobbyists have a zero temperature coefficient, which is not what is needed.
Possibly a "huff and puff" stabiliser circuit would be the answer. Unfortunately I'm not an electronics designer so coming up with a design for one would be beyond me. Cumbria Designs has a kit for a board called the X-Lock VFO Stabilizer which might do the trick, but I can't find anything on their website that would suggest how it could be incorporated. I'd have to lose the internal battery to make room for it, but if it could solve the problem then it would be worth it.
Since the Cub starts to drift after you start to transmit, even when it has reached a stable operating point on receive, it's clear that the small amount of heat radiated by the PA transistor heatsink is causing the problem. My latest idea is to try to prevent the heat from getting to the VFO components by making a screen of thin card and fitting it across the Cub circuit board so the PA is in one compartment and the VFO in another. This hasn't eliminated the drift but it seems to have reduced it a bit further, which is at least a step in the right direction.