A gentle reminder that the Korg Kronos -- which is, basically, every kind of synthesis you could ever want, in one keyboard, the last of the Great Workstations -- is, ultimately, a pretty old and slow PC running Linux integrated with the interface.
Any i7-based (aka "modern" since around 2011) PC is multiples faster than that.
The bias towards any hardware digital over software in terms of sonic quality is purely arbitrary, and psychological, definitively NOT technical, well except maybe in the sense that older technologies, because familiar, sometimes catch the ear in distinct ways that due to habit are considered "pleasing."
This debate went on endlessly in the then-existent Clavia Nord Modular forum, when the G2 came out and the G1 wasn't so old most people had forgotten about it. Long-seasoned G1 experts, like Rob Hordijk (inventor now of the Rungler, among other fascinating modular contraptions!), Chet Singer, Marko Ettlich, a host of others of that level of expertise and musical sensitivity, almost universally complained that the G2 didn't sound "as good as" the G1. Possibly because it was using newer-generation DSPs, instead of Ye Olde Motorola 56303s, which everyone knows have as classic a digital sound as Curtis chips do analogue?....

[the 56303s are in virtually every digital synth made between '96-2002, well except for the larger Japanese companies using proprietary self-created hardware or, honestly, early '90's Roland stuff which -- I kid you not! -- used the same CPU used in the Sega Genesis game system!].
That anorak/grognard rejection of the G2 basically helped kill the market for any future Nord Modulars, I suspect. Not intentionally -- there was/is a difference in sound, due in part to the hardware, in part to the increase in quality and precision of the technology available, in part due to Clavia's research pursuit of advanced FM possibilities with the G2 instead of simply recreating past success.... lots of factors.
ANYWAYS, the point is mainly that there ARE differences in different digital hardware, but almost any even half-competent software made these days is light-years ahead, in terms of both resources available and software circuitry design subtleties, of anything made even three-four years ago.
And, ultimately, if it's digital.... it's all software.
A/D-D/A converters have -- even the cheapest -- long ago achieved "Burr Brown" levels of quality, to the point that they're not a factor in sound any more. I suspect a lot of people use crap sound cards in their PCs, or have lousy monitoring, and thus judge the output accordingly.
Any sampling rate above 44.1kHz only matters in terms of the internal mathematics of digital circuit processing; if you have a lot of numbers at high speed getting passed through a lot of calculations at a level of multizillions a nanosecond, small losses in digital precision can add up and produce barely-detectable noise or coloration in the final output -- I'm talking internally in the software calculation process long before anything goes to an output of the software circuit -- and that's why using 48kHz or 96kHz oversampling can produce detectable sonic differences. But in terms of real-world needs? Except for pseudo-scientific GS "findings," irrelevant, even there.
Reaktor Blocks tends to be very CPU-intensive because I believe they were originally built with Core, which as has been stated, is not multi-threaded in its design, would be/is (no idea where NI is in thinking about doing this) an absolute monster to redesign to be multi-threaded and use multiple processors, and is at root trading off computational efficiency for flexibility and exposure of things down to the z-1 sample level for DSP circuit design emulation (without needing reference to SPICE) available to end users. The number of people on the planet interested and competent enough to actually use Core effectively could probably fit comfortably into a studio apartment in downtown Berlin, so it's not a very profitable idea to think about spending lots of development time on making things multi-threaded for those six people.....
Open up Razor sometime in a full version of Reaktor, and drill down to all the component pieces. You will cry. It makes the average circuit design for a synth look like child's play by comparison. Any of NI's softsynths released since Razor tend to have similar levels of complexity; a lot of knowledge and insight and foundation has been built up over twenty years within the Reaktor community and library, and in NI's own independent research, that provides the basis for more recent softsynth output.
BTW.... inside NI's Monark, my fave name for a module of ALL TIME! is the "Mix&BlubBlub Enhancer" in the circuit.
I think that's the magic of Moog; it's Blub Blub Enhancement, clearly!
So.... there's all that.