Except you can't just cram more transistors on a chip and make it faster. There are practical limits to how many transistors you can put on a single chip/how big you can make chips, and the M1 Ultra (if you count it as a single chip) and M2 Max are already well ahead in the "how many transistors can you put on a chip" race (
https://en.wikipedia.org/wiki/Transistor_count).
Maybe you can squeeze in more
cores but then you're up against the limit of how efficient software is at parallel processing. Or you can add specialised "engines" for things like machine learning, encryption, media codecs - which is exactly what Apple have
already done with the M1/M2 - but, again, that requires software to take advantage of it. Even
supercomputers tend to be boring old PPC or ARM cores driving banks of specialised vector processing units (or consumer GPUs in economy class) that provide the "super" bit - for specially written software.
One problem with the Mac Pro customer base is that it includes a
lot of customers who are only still using MacOS because it would be too expensive for them to make major changes in the applications they use. A courageous new massively parallel system that relies on software being re-written is
not necessarily going to fit the bill.
Anyway, I think you're missing the point that the M1 Ultra
already provides enough CPU power to drive a credible Mac Pro - the issue is that it can't provide the
expansion possibilities which are all that distinguished the 2019 Mac Pro from the iMac and trashcan. it's designed from the ground up to use tech like unified RAM, integrated tile-based GPUs and Thunderbolt 4 for expansion, and if Apple wanted a new Mac Pro in the same mould as the 2019 MP they'd need a totally new chip.
Designing courageous new housings for Intel chips is small beer compared to the cost of developing a new Xeon/Threadripper-killing Apple Silicon CPU
die just to serve Mac Pro customers (c.f. the M2 Pro/Max/Ultra which re-use the same M2 Max die design for everything from a 14" MBP upwards). The cost of making processor chips is hugely dependent on economies of scale. The 28 core 'M-suffix' Xeon W chip retails at about $7000, and
that's paying 2x the price of the non-M-suffix version just to get support for an extra 1TB of RAM (those
darn Safari tabs!). That's a product with a far bigger potential market than just a single, and relatively low-selling, Mac model.
Speed-bumping the 2010 Mac Pro for 10 years may seem like the better decision in hindsight - but it would have played to an ever-shrinking pool of customers as pro users gradually jumped ship to cheaper but equally capable Intel/AMD gear. The Trashcan was an attempt to offer something distinctive - mainly, they made the wrong call on the 1 CPU + 2xGPU configuration. The 2019 MP - well, YMMV, personally I think it was a hideously over-engineered parody of the 2010 Mac Pro designed-up to a ludicrous price, but OTOH it was a step forward for Mac in terms of the level of expandability and power available (at a price). OTOH that was mainly just because it was an early adopter for Intel's new Xeon-W chip which had the necessary PCIe and RAM capacity. More powerful PC workstations are available if you look beyond HP's consumer website.