At high frequencies (anything that isn't 32.768 kHz) the resistor in parallel with a fundamental crystal does not need to be a precise value.
It is only there to bias the amplifier to a mid point for start-up. Without the resistor, the input will float and it is unlikely to start, but once it is running the resistor is not needed.
As long as the impedance of the resistor is large compared to the capacitors at the operating frequency, and small compared to the leakage current, it will work.
At 25 MHz, the impedance of a 10 pF capacitor is about 600 Ω. The leakage on a CMOS gate is hard to measure, but obviously the 10 MΩ resistor works. However I think that anything in the 10 kΩ to 10 MΩ range would be fine.
If the original problem was an error of seconds per minute, that is far to far to be adjusted for by altering the capacitors on a crystal. That would only change by 0.01 % or so. Errors of a second every few hours can be adjusted by altering the capacitors.