My own 2 cents on this; it really (on Windows) is a case by case thing. My simple test is to pick an average session(for me) and run at buffer settings of 64 and 1024 and note the system usage cpu(total, not each core) with hyperthreading ON. Then repeat with hyperthreading OFF. The numbers tell the story. If they are significantly better, either way, I will run the way that reports the lower cpu usage(unless it suffers stops or error messages). If I see no difference, I tend to leave it ON. In the past, some users with very similar computers would have very different results(why I feel its a case by case thing).
FWIW, 2019.5 seems to show insignificant changes in cpu usage with HP on vs off, but the good news(for me, anyway) is that 2019.5 is way easier on cpu than any previous build. I am now running sessions at a 64 buffer that would never run before without stops and 9xxx errors(even with Melodyne and Ozone 8 in play)