It’s based on overall “ASIO performance” which is based on Cubase’s ability to “schedule audio packets” across available CPU cores.
Enabling ASIO Guard helps in most cases. ASIO Guard is similar (speculation) to Logic’s “hybrid” audio engine. Basically, some sort of pre-rendering happens to tracks that aren’t “live / real-time.”
Increasing the audio interface’s buffer is the main way to improve this. Latency is the tradeoff.
The job a “audio buffer (packet) scheduling” is a complex one that all DAWs do slightly differently. It’s a tradeoff of raw latency vs. ability to splits up the job across more cores.
Factors in the signal chain such as “serial” vs “parallel” chains, factor in. Live effects on the stereo bus are usually examples of the “serial chain” and tend to be the hardest to split up across cores, for most DAWs. Reaper being the exception.
Individual VSTi channels, that feed into the stereo bus, are often examples of “parallel” chains for some DAW (notably, Sonar). They can often be split across cores more easily (again, Sonar works this way).
The Reaper DAW is an example of a DAW that does the most aggressive “slicing up” of the scheduled audio packets across all cores, and is why it will utilize the most of a CPU of all the DAWs (and therefore most closely matches the Task Manager meters).
Cubase, especially with the new ASIO Guard, probably comes in second place.
As ASIO Guard improves, the VST performance meter will start to more closely match the Windows Task Manager meter.
The “audio-driver” and the audio interface has also impact on the meter inside Cubase. Change to another audio interface and vendor can give you different results too.
I’m also still looking for a good explanation of how the audio engine distributes the workload onto the different CPU threads and what the “average load” and “real time peak” represent.
For example, it doesn’t make sense to me (as yet) why the “average load” is always higher than the “real time peak”. The former is supposed (by definition) to be somewhere between the real time peak load(s). Is the “average load” not an “average” “load”?
I’d very much welcome a good explanation of VST Performance metering from Cubase’s engineering team.