This line of reasoning feels unconvincing. If there were really worried about this they should disable it everywhere.
So: cloud computing, and consumers that might be running dumb code from the internet (= i{3,5,7} and Xeon = volume production) should get chips with no hyperthreading; while HPC, locked-down workstations, and gaming PCs that probably don't have any important data on them (= i9 = low-batch-size production) should get chips with hyperthreading.
For HPC it's usually turned off, and I think for gaming and workstations it also has little if any use.
Also, HPC doesn't necessarily imply a uniform workload; it just implies that all workloads are trusted. Really, modern HPC clusters are still "multi-tenant"; they're just tenants who are all e.g. employees of the same company, or researchers at the same university. So you can describe them, from a security perspective, as one "tenant." None of them would have anything to gain by doing a side-channel attack against one-another.