The hypothesis is that txs fees size scale with fiat-value of bitcoin. And it makes sense.
Money that competes for blockspace is mostly denominated in fiat.
Miner costs are denominated with fiat. Opportunity cost of using bitcoin is also denominated in fiat. Meaning
Think about it. Bitcoin network competes with meatspace entities:
  • If it costs me 1$ to send 10000$ of remittance with bitcoin I will use bitcoin.
  • If it costs me 1000$ to send 10000$ of remittance with bitcoin I will use western union.
If most economic decisions of "should I used bitcoin or $SOME_MEATSPACE_PRODUCT" are denominated in fiat, the competition for blockspace will be a function of fiat-denominated value (and not bitcoin-denominated value).
There are cases in which bitcoin have no meatspace competition (like censorship-resistant money) and those cases might cause a decoupling. But arguably most competition for blockspace at the moment are driven by competition-with-meatspace-entities.