Jfets are horrible for repeatability. Idss specs can be 3:1 if you're
lucky, 10:1 on some parts. Enhancement mosfets won't be worse that
that. But any fet-based open-loop attenuator is going to be bad. If
you close a loop around it, an enhancement device, run at low voltage
swings, will work about as well as a jfet.
If you don't care about the noise (times if not orders of magnitude
higher) then a 2N7002 or sort of might do.
Repeatability is not that bad, with the 2N4391 you can rely
on it going down to around 30 Ohm at 0V GS over temp (some will go
well below 20 Ohm at 25C). I know this because I have done variable
gain using FETs 15+ years ago; I had to pair them (no paired 4391,
only 4393 as 2N5566 IIRC) for temperature compensation, to eeprom
calibrate against temperature drift (which was still too high
for a 13 bit ADC), and to digitally linerarize the conversion
results (12 bit INL was OK but with FETs you go into a few % INL,
1% is the best case)...
- photo taken with the 2N5566,
before the paired 4391 were bulkier and uglier (soldered cans,
- the companion board,
- the servo driven rotary switches,
caps etc. not yet soldered when photo was taken,
- boards & servo put together,
- the still kept historic reference
to the entire mid-90s designed madness).
But since he cannot afford the -10V headroom a 4391 would take
this is a non-issue, either the 2N7002 will do or not.