I might guess your using spread spectrum. Intel default is usually 0.5% down spread so with 100MHz BCLK that would have the clock changing from 99.5MHz to 100MHz and back again linearly. This is done to reduce EMI as having the clock in one place will produce a higher signal of interference than spreading it over a band of frequencies. This means your average clock would be about 99.75MHz (might actually be closer to 97.76MHz). Your screenshot shows FSB
RAM as 24:1 (not sure why using old FSB instead of BCLK) so 24 x 99.75 = 2394MHz. If that's so then it's really good as software measurements can cause differences too. On my HW CPU-Z use to show BCLK bouncing by a MHz or more but was improved a lot some time ago.
As for drift it is possible for the reference crystal to significantly change frequency with temperature but because all clocks can be affected you might not see any difference with software derived clocks. I have seen this on my own hardware using an external frequency counter connected to PCIe slot clock (PCIe tied to BCLK). I wonder if they actually check this when trying for LN2 OC world records otherwise results could be false.
TLDR Try disabling BCLK spread spectrum if possible and set BCLK to 100.0MHz.