This really requires a book, but here goes:
The shunt capacitance of a crystal is easy -- that's just the capacitance due to the quartz as a dielectric and the electrodes as plates. It's the capacitance of the crystal when measured at low frequencies.
The load capacitance is the capacitance you need to present to the crystal to make it resonate at the desired frequency. If you just had a crystal and no amplifier you'd just put that across the crystal:
+---||---+ | | ---+--|X|---+---Now if that's 20pF and you split it into two capacitors in series, you'll need 40pF caps:
+---||---||---+ | | ---+-----|X|-----+---Now if you ground the node between the caps you'll get:
---+----|X|----+--- | | === === | | V V
The crystal doesn't know the difference. Now if you hang an amplifier between the two ends of the crystal you'll get:
| +-| >-+ | |/ | | | +-+-|X|-+-+ | | === === | | V V
If the amplifier is a perfect transconductance amplifier with infinite input and output impedance then the crystal will resonate at the same frequency and the output will grow until it stabilizes when the amplifier output limits.
OK, that's cool -- but a real amplifier has some resistance and capacitance at both input and output. If the amplifier will work it'll have enough gain to overcome the input and output resistances. To make the oscillation frequency right _you_ have to subtract the amplifier's input and output capacitances from your 40pF capacitors -- and that's why you see lower capacitances at those spots in real circuits.