I've seen scenarios where HT doesn't help, iirc very CPU-heavy things without much waiting on memory access. Which makes sense because the vcores are sharing the ALU.

Also have seen it disabled in academic settings where they want consistent performance when benchmarking stuff.