I should have been more clear that the above was an attempt to set crude lower bound based on the naive code I had at hand to illustrate that 8 ksps was more than plausible.
I was a bit surprised there's that much headroom and will have to look closer at your example, but much of the difference might be down to the "analogRead()" on the STM32 being horribly slow.
Apparently I'll also have to figure out how to slip an R4 purchase by MrsMark.