As it turns out, Bech32 only represents 3.5% of the network if measured by number of UTXOs, or 4.6% if measured by the number of BTC stored. From a privacy perspective, it seems reasonable for ChipMixer to continue holding off.
Segwit adoption is quite larger than this.
I made this topic about it, where I made a few charts.
I used blockchair data from LoyceV csv file.
This chart is the witness_count/transaction_count ratio.
As witness_count is the number of transactions in the block containing witness information, we can easily calculate the segwit adoption per block.
As the chart would become too noise with adoption per block, I calculate the adoption per month (witness_count per month / transaction count per month)
This is the result. Nearly 50% of transactions are segwit.
As 50% of transactions are segwit, ChipMixer would be more private if nobody knows which address format it is using. For now, everyone can know for sure that 50% of the transactions are not CM chips.