One problem with the 6502 was the tiny 64k address space, the other was that the 6502 was a terrible compiler target with the result that compiled languages for the 6502 usually used virtual machine techniques that gave awful performance like the atrocious UCSD p-System.

There was the 65816 which was clocked higher and had a bigger address space but did nothing for the compiler problem and did not have 24 bit index registers to go with the bigger address space and didn’t have anything like the segments in the 8088/86 that let you do pretty well despite not having full size index registers. In an alternate universe there could have been a path to 24 bits (80286) and then 32 bits (80386) that was compatible with the 6502 but there wasn’t.

The Apple //gs was an impressive machine that looked good compared to the very expensive Mac 2 and I think that’s what your ‘Super 128’ might have been at best. Have you seen

https://www.commanderx16.com/

? I also think the only 24-bit extension of an classic CPU that I like is

https://en.wikipedia.org/wiki/Zilog_eZ80

which really has 24 bit arithmetic and index registers and is based on an architecture which is compiler friendly.

GEOS was also released for the Apple II and PC/GEOS for IBM PC compatibles in 1988 and 1990 respectively, even beating Windows 3.0 to the market. Neither version established any substantial market share.

There simply was never substantial demand for GEOS. While actual sales figures are hard to come by, it seems very likely there was substantially less interest in GEOS than in the Amiga platform. It would not have "saved" Commodore.

Commodore needed more memory. We already had megabyte machines in the late 80's.