The „stressometer“ indicates the complexity or „stress factor“ of the patch, not the current CPU load. A patch with level 1/5 can be played with much higher polyphony (something like 90 or so, we did some tests here on the forum) than a patch with level 5/5 (max polyphony about 6) before voice stealing will occur. Also, the Virus will apply several strategies before starting to steal voices, such as temporal „smearing“ where chords are audibly arpeggiated, envelopes become less snappy and LFOs fall out of sync, as well as audible degradation of complex sounds by reducing the harmonic spectrum which is most obvious on FM sounds.
Also important to know is that voice stealing and graceful degradation are handled differently in MULTI and SEQ modes. It seems that in SEQ mode, Part 1 can use both DSPs while all other Parts are only using one DSP with all odd numbered share one DSP and all even numbered Parts share the other DSP. This knowledge can come in handy to optimize distribution of load for an arrangement.
Useful info Oli.
I've also got my 3rd virus en-route (a white Polar) so with 3 x Virus' on my A-Frame - TI-61, Darkstar and White Polar (goodbye Arturia Minibrute heap of crap) I'm sure I'll have enough parts / voices / processing power !!! Can't get enough of the virus and I'm only at the start of my journey.