That is just the nature of CPUs (DSPs) reaching their processing limit. Everything is always calculated in sequence and not in parallel. The more complex a sound, the more time it takes to calculate the waveform. At some point, you have to either defer some calculations or simplify them by degrading the audio signal. Both will happen on the Virus, with sound quality as priority. It seems that timing of Envelopes and LFOs is impacted first; I observed this behavior accidently when I had programmed a very short high pitched FM based tick sound, which became considerably "broader" and "imprecise" as more other sounds (basses and pads) where added.
Earlier TI OS Versions had substantially less Oscillator complex types and Effects. I guess that has more to do with processing timing behavior than potential changes of the MIDI over USB timing.
The example I mentioned is a patch OS 3.3.4 can identically make. My point is that midi timing of the exact same patch will remain solid in 3.3.4, even upping the unison to x8, while in 5.1.7 the timing starts being off even at 2x unison. Sure, because it's a heavy patch (intentionally so ), voice stealing will occur in both OS-ses, but in 3.3.4 voice stealing doesn't change the timing of the notes played, while in 5.1.7 it does.