Wirth decided it would be better to make a simple, custom processor that follows the spirit of the system. That by itself would be a fun project. He could put it on any current and future FPGA, something very unlikely to disappear. He then ported Oberon to it. Now, it's future-proof.
Oberon-related tech is also training for ETH students. They get plenty of experience trying to extend the OS, port the compilers, etc. They have little in hardware by comparison, though. FPGA's have gotten cheap and their nature means they'll stay around in some form forever. Verilog also goes back decades. So, there's the added benefit of teaching students to build a processor which could also run the system.
So, combining future-proofing, simplicity, OS education, and HW education requirements all into one meant a custom, Wirth-style processor on FPGA was best option. It also creates opportunities for things like TCP/IP, graphics, or security engines he probably didn't think about at the time. His students or FOSS volunteers might pick that up given how simple the HW is.
So, ARM is a bad idea. Those SOC's are too complex. Extending or re-implementing it needs a ridiculously-expensive license. They also sue people over even using the ISA, which makes me boycott them where possible. The FPGA doesn't have these problems plus allowed he and his students to attempt an ideal replacement that could last decades. So, that was his choice and one of the better options.
Note: I'd have used it as an excuse to get some grant money to do a full, low-cost, RISC-V implementation to put A2 Bluebottle on. More practical. That's just me, though. ;)
What I would like to see instead would be RISC-V instead of their own ISA. At least there some learnings could be shared with other academics and tinkerers.
Note: Areas like embedded in general with smaller, custom jobs don't worry about ISA's as much. Plenty of ARM, MIPS, PPC, x86, SPARC, Super-H, 16bits, 8bits... you name it. Much more fun space to be a programmer if you get to pick the cpu/board. :)
So, I do not see any reason for the Oberon machine to rely on any "standard" ISA. It can even be compiled into some kind of a NISC, and nobody would notice any difference.
Now, if we're talking GPU's, the effective ISA would be DirectX, OpenGL, OpenCL, etc. They're the standards that software and tooling target. So, does you product get by with not supporting any of those? Or do you have to comply with the your niche's standard interfaces and ecosystems, too?
However, my usual recommendation is SPARC given its ecosystem, openness, and mere $99 registration free. RISC-V, esp Rocket processor, will hopefully take its place. A custom, tiny RISC chip on a FPGA? Ok, ok, Wirth... that is more practical... (sighs) Lol...