Hyperthreading works really well when the same code is running on each core, but the data dependencies are different. Databases and ray tracing are probably the two best examples of this, but any parallel code that does a lot of random memory access or "pointer chasing" will benefit a lot.

With typical database servers, hyperthreading provides about 1.4x the performance "for free" because it allows execution units to do something useful while another thread is waiting for memory.

Where it isn't useful is typical game engines, except in some corner-cases as you've noticed.

Typical game engines like to have dedicated cores with the highest possible performance. Hyperthreading improves overall throughput at the expense of single-threaded throughput.

Similarly, game engines tend to run task-per-thread with different code, which tends to pollute the L1 code cache too much. So if two different threads are running on the same core but are doing different things, they'll fight over this precious resource.

Intel measured something like a +20% overall improvement averaged out across a wide range of workloads. A few had regressions, and a few (like ray tracing) had huge benefits.

IMHO, hyperthreading makes the most sense for servers, or laptops with a small number of cores (2-4). For high-end gaming, 16 cores with HT off will likely have the best performance.