Ken Scharf wrote:

Since all the frequencies in question are harmonics of 3.5mhz
(except 3.3mhz) you could use a PLL to generate them by phase
locking a vco to the harmonics of a 3.5mhz oscillator.
Use a divide by N circuit to divide down the vco output to
3.5mhz and have a phase detector compare the output of the
divider with the 3.5mhz oscillator. Properly shielded,
this pll would be as clean as the crystal oscillator.

Since they are low order odd harmonics, is a PLL even necessary, or
could a square wave sort of buffer, a filter at the desired frequency
and then an amplifier tuned there do the trick?

With a PLL, you will still need to switch components for each frequency
range - generally you get less than an octave (factor of two in
frequency) out of a single VCO, but changing the inductor and/or fixed
capacitors for each band shouldn't be too hard.