I recall that the perf approach to off-cpu flamegraphs can be problematic because of the cpu time overhead and because some of the command line arguments involved have changed over kernel versions.
I can’t personally give a blanket recommendation about hyperthreading. Testing it with your workload is a good idea.
I do want to note, however, that off-cpu flamegraphs are probably not relevant to hyperthreading. Hyperthreading will affect the throughput of threads that are running (and sharing the resources of a single CPU), but shouldn’t directly affect how much time they spend in a non-running state, which is what off-cpu flamegraphs measure.