Performance scaling across cores

I can't speak for Windows or Linux users but on Mac I have two scenarios with the standalone version that make me think that 4 cores (8 cores hyperthreads) is a good combination. On a 2013 Mac Pro with latency 1.3 ms (64 samples at 48KHz) only 4 cores are ever used for a normal preset and 16 threads are running. On a 2019 MacBook Air with a 2 cores (4 hyperthreads) all 4 cores are used all the time. Only when I run a layer preset (with the maximum 3 instruments) do all 8 cores on the MacPro show (mostly even) activity and the thread count goes to 19. Organteq in comparison, always uses all 8 cores on my Mac Pro.

I'm sure using multiple instances of Pianoteq in a DAW can benefit from more cores though.

I assume that by voice you mean a preset. As far as I know, the polyphony setting corresponds to the number of notes of the preset allowed to play at the same time, these notes of course have many harmonics to be computed and interact with currently sounding ones. I usually set polyphony to 256 and, even on my MacBook Air dual core i5 with slightly higher latency (128 samples, 2.7 ms) keeping the sustain down and running fingers up and down the keyboard until I reach 256 doesn't result in crackles. I think recent cpus, even not the fastest ones, will run Pianoteq standalone without any restriction to your musicality.

Rendering dense MIDI files of Liszt or Rachmaninoff, I only noticed surpassing 256 when using the 3 layer preset I posted about somewhere else. That was the equivalent of 3 pianos playing the same piece, and then using Auto polyphony is necessary to avoid overload.

Regarding latency, I'm referring to the value showed in Pianoteq Options. As was discussed somewhere else, the final total latency including the whole chain of events may be larger. In my case, I don't notice any latency with these settings of 64 or 128 samples.

Forgive me if this is irrelevant on Macintosh, but... the audio interface, its drivers and/or its communication channel to the computer?

At least on Windows, with ASIO, the buffer size in the Pianoteq options is the buffer size in the audio interface drivers.  I suspect the interface, its drivers and the communication channel (e.g., USB, Firewire, Thunderbolt and the associated chips and drivers) are a significant factor once the CPU is sufficiently powerful that Pianoteq itself isn't overloading the system.

If you two are using different audio interfaces, it might be those -- and not the CPUs or core counts -- that are making the difference.

Looks like something Apple Silicon based (pending a native update) is going to be a winner in that case. Perhaps I'll get a Mac Mini to dedicate to Pianoteq duty.

It would be nice to have some sort of little benchmarking tool for Pianoteq - the 'performance index' doesn't really give me anything terribly meaningful.