I'll try and draw what I have in mind, and post it in alt.binaries.schematics.electronic. However, in the meantime, let me try and explain. The explanation may not be absolutely 100% complete, or even 100% correct, but it may help in moving a crystal more HF than it wants to go. Sorry that it's a bit rambling!
In Anthony's circuit
As Steve has stated, a crystal suddenly goes into parallel resonance just HF of its series resonance. This limits how far the series resonance can be pulled HF by the addition of a series trimmer capacitor. However, if this parallel resonance can be removed (or moved further HF), it should be possible to move the crystal further HF. The technique described certainly does work with VHF overtone crystals (between 50 and 200MHz), but should also work with HF crystals working on their fundamental frequencies.
A crystal is a mechanical device, but can be represented as being a series-tuned L-C circuit. (Call these L1 and C1.) Also, across the two is a parallel C (C2). Forget about losses (represented by a resistor). [Note: L1 and C1 are not actual electrical components, and only appear to have these values at or near to the L1-C1 resonant frequency. However, C2 essentially is a physical electrical capacitor consisting of the plating on each face of the crystal, with the crystal as the dielectric between.]
L1 is very large (possibly 1H or more, depending on the frequency). C1 is very small (say only a few pF or even a fraction of a pF - again depending on the crystal frequency). [So adding a relatively large series trimmer capacitor has very little effect on the frequency.] C2 is typically around 5pF, regardless of frequency.
Imagine doing a test where you look at the resonant frequency of a crystal, using a signal generator. This feeds an RF signal through a crystal, into a 50 ohm load. You measure throughput of the crystal by measuring the voltage across the load.
Swing the sig gen frequency slowly from LF to HF, through the resonant frequency of L1-C1. [Let's forget about C2 for the moment.] Below the resonant frequency of L1-C1, the L1-C1 circuit acts like a small capacitor, so there is very little throughput. Above the series resonant frequency of L1-C1, the L1-C1 circuit acts like a large inductor, so again there is very little throughput. However, when you hit the series resonance of L1 and C1 (F1), reactance of L1 and C1 cancel. The crystal acts like a short-circuit (or nearly so) and there is a large throughput. Because the L-C ratio is very high, the resonance peak is very sharp.
The effect of C2 across the L1-C1 circuit is to produce a second (parallel) resonant circuit. VERY slightly HF of the L1-C1 resonance, C2 resonates with effective inductance of the L1-C1 circuit. This produces a parallel resonant circuit (F2). Another way of looking at it is that L1 resonates with the series combination of C1 and C2 (so F2 must be higher than F1). The parallel resonance is, of course, a high impedance, where there is almost no throughput through the crystal.
As a result of this double resonance, the crystal acts as a series-tuned circuit at F1 (one you want), and a parallel-tuned circuit at F2. The transition between the two is very sudden. The frequency response peak of the throughput is very lopsided, and gets chopped off suddenly on the HF side.
The difference between F1 and F2 is very small (a few Hz to a few kHz, depending on the frequency and type of the crystal). If F1 is lower than you want, and you add an external series trimmer capacitor to try and pull the crystal L1-C1 series resonance HF, you effectively hit a brick wall with the parallel resonance at F2. The parallel resonance will block any throughput at (or near) this frequency.
A possible solution is to neutralize C2. [Note: Neutralization is a technique sometimes required when using VHF crystals, as C2 may be large enough to allow the oscillator to free-run, instead of being locked to the frequency of L1-C1. However, it may also be used with advantage, as described below.] You can neutralize C2 by adding an inductor across the crystal (ie in parallel with C2). The value required is that which parallel-resonates with C2 at the crystal frequency. In effect, C2 no longer exists. With C2 neutralized, there is no longer a sudden transition from the wanted series resonance F1 to the unwanted parallel resonance F2. The peak the response curve of the throughput of the crystal (at F1) is now nice and symmetrical, without the sudden cutoff at F2. In practice, the actual F1 peak will probably be somewhat more HF than before, and the crystal should be more pullable with a series capacitor.
Finally, if you reduce the value of the inductor so that its resonance with C2 is somewhat higher than the crystal frequency, this tends to pull the F1 resonance peak even higher in frequency. However, if you overdo this, the oscillation will probably unlock from the crystal, and start to free-run.
As I said, sorry for the ramble. Ian.