The care and feeding of oscillator crystals is a whole science in itself. I tried to go there but, stopped because it is too vast for me. Besides with all the stray capacitances we add when building our PCBs and breadboards, it ends up to be 'cut and try' anyway.
The load capacitance is the correlation value that the crystal manufacturer used to bring the crystal to the specified frequency. Typical circuits use two caps and they do not have to be equal in value. In fact, one can be fixed and one can be variable so the crystal can be tuned to exact frequency. Parallel resonant units, like we use, want to run at with 180° phase shift in circuit. (Series resonant units like zero phase shift, so no load capacitance is stated for them.)
Electrical Engineering students understand why crystals are used instead of capacitors as the frequency-determining element in an oscillator. A crystal is made from quartz (natural stone), whose electrical equivalent circuit is shown in Fig-1. The key characteristic of crystals is that their parameter values remain stable over time, temperature, and humidity.
In contrast, capacitors are made from organic materials like paper, oil, and foil. Their circuit equivalance (Fig-2) is also similar to Fig-1, but their parameter values can vary significantly with time, temperature, and humidity.
Despite this, we add 22 pF/50V capacitors to the legs of the crystal. The reason for this has been explained well in post #82 @herbschwarz. A similar explanation can be found on Google:
"The capacitors placed on either side of the crystal and connected to ground provide the proper phase shift around the closed loop network. This ensures that the gate input is in phase with the gate output (Fig-3), a condition necessary for sustained oscillation."
Figure-1:
Figure-2:
Figure-3:
The issue to be discussed here is why 22 pF/1 kV capacitors did not work for the OP.
The following explanation is found in an online document:
"Parasitic Elements: Higher voltage capacitors might have different parasitic elements (like equivalent series resistance (ESR, Fig-2) and equivalent series inductance (ESL, Fig-2)) compared to lower voltage capacitors, but for small capacitance values like 22 pF, this difference is usually negligible for most practical purposes."
Tom.. 
220M is 220pF
They are not.
Old stock high voltage caps could be marked with a straight value, so 220pF.
Leo..
Accordiong to above Table of post #26, 220M stands for:
==> 220M
==> 22 x 100 +/- 0.20 x 220 pF
==> 22 x 1 +/- 0.20 x 22 pF
==> 22 +/- 4.4 PF
==> 17.6 pF to 26.4 pF
The upper value of the 220M capacitor has exceeded the recommended value as is seen in the following Table of Atmega38P data sheets.
Why does 22 pF/50V capa work?
This capa has the following tolerance which is muh better than 220M/1kV capa.
There are exceptions to every "rule".
'102' is 1000p, '101' is 100p (it won't be 101p).
I've likely never seen a cap < 100 marked with a 3-number 'code', just the value with an underline or a black tick above.
PE - I have some 1KV Z5U's, look just like the OP's, and they're marked 4700M - and they're 4700pF (0.0047uF).
So, the best practice is to go with the manufacturer's specifications.
The ule of thum is:
Values in pF are shown as naked numerals like 104; where the last digit is the power of 10.
==> 104
==> 10 x 104 pF
==> 10 x 10 x 104 x 10-12 F (farad)
==> 10 x 10-8 F
==> 0.1 x 10-6 F
==> 0.1 uF
This, for sure, is for power line applicationa and NOT for low voltage electronics.
Well, see Post No.77
