Here is what I typically use. The bias values for the
resistors on Q1 may be off and have to be adjusted.
Working from a 62 year old memory here.
Q2 is a buffer (emitter follower) which minimizes the
connections downstream reflecting their load back
to the oscillator and causing frequency pulling.
I usually do not use a resistor in the collector of
the buffer or oscillator. The Colpitts as shown is
an emitter follower which provides POWER gain.
For a LC tank I usually set the tap at 50%. Thus
if you have 100 pF as the resonating cap then use
200 pF for each. Similarly if your resonating cap
is 50 pF then use 100 pF for each.
For crystals I start out with a 1:10 ratio. The
objective is to use as small amount of feedback
as possible to get the circuit started and sustain
operation over the temperature and operating
voltage range. The less feedback through the
crystal the less heating and less drift. (Yes even
the small microwatt feedback levels on quartz
create enough heating to shift the quartz
frequency of resonance).
As for getting better than the 1 MHz you are
currently experiencing over 15 minutes you are
doing pretty darn well already. LC tanks can be
fairly stable over temp with tolerable drift up to
5 to 10 MHz (my personal experience). At 100
MHz you really need either a quartz resonator
or a PLL system otherwise the drift is likely to
be way more than you want to tolerate. Radios
from the 50's through the 70's used phase locking
techniques to keep the receiver LO locked to the
incoming station the listener tuned to.
If you use a crystal you really need an overtone
rock. 5th or 7th overtone will get you between
the 100 and 200 MHz range. Don't try to use a
fundamental crystal in an overtone application.
The harmonic resonances in fundamental crystals
are much lower amplitude than the fundamental.
You will find yourself chasing your tail trying
to add selectivity in the oscillator to narrow
the amplification bandwidth so the other resonances
do not pull you off frequency. Also the harmonic
resonances are often not exact multiples of the
fundamental. They will be close but off just
enough to be maddening.
Hope this helps.