Polystirene (styroflex) capacitors are a good choice. They have a
desirable slightly negative temperature coefficient, and excellent Q.
1000pF is within the spectrum of usual values.
Or you may choose a bunch of NP0 ceramics in parallel, plus ~1/20th of
the total value made up of a N750 (-750ppm/centigrade) capacitor. Many
paralleled caps are better than a single multilayer unit because heat
(yes, real life caps do heat somewhat in oscillators) is better
dissipated away, thus reducing one of the drift causes.
73s and DXs
Daniel Perez LW1ECP
==================================
I support the above opinion ,having made a simple FET based vfo for
3.5-38 MHz which serves as a training tool for the UK Intermediate
Licence practical assessment ,for which trainees have to calibrate a vfo
against a receiver and/or frequency counter.

The VFO was made (semi-ugly style) on PCB board with 'ground' holes
drilled for upright grounded components ,with the other lead to support
the relevant non-grounded components resulting in the shortest possible
interconnections.
Using polystyrene caps and with the VFO not having any buffer stage the
device fed by a 9v PP battery drifts not more than 20 Herz in 30 minutes.

Frank GM0CSZ / KN6WH