You can't compare clock speeds in that way and get an accurate picture on performance. If you could, then a 3.8GHz Pentium 4 would be faster than a 2.8 GHz Kaby Lake Core i7.
To give a better (but still not completely accurate) idea of how far apart the SNES and N64 were in power, look at the difference in MIPS (Millions of Instructions Per Second) in just their CPUs:
SNES CPU (Ricoh 5A22): 1.5 MIPS
https://en.wikipedia.org/wiki/Ricoh_5A22
N64 CPU (NEC VR4300): 125 MIPS
https://en.wikipedia.org/wiki/Nintendo_64_technical_specific...
And how could it not be? The two systems were released 6 years apart (if you count SNES in Japan 1990), back when Moore's law was in full effect.
The SNES was a 3.5 MHz 16 bitter. The n64 was a 3d-capable machine with a 100ish MHz R4k, the RCP, an analog controller, etc, in the box. The advance just seems enormous (matching similar advances in personal computers in the same timeframe). If I'm understanding you right, you're making a case the advance was maybe not as discontinuous. But the scale of it still seems qualitatively greater than any of the subsequent ones, to me.
Now, given that the SuperFX's lifetime was short and its capabilities were likely never fully explored/exploited in that lifetime; and given that the SNES's on-cart coprocessors could completely override the console processor and just bang bits onto the video-output lines themselves (like the Super Gameboy does!); and given that the only thing stopping devs from sticking even more powerful chips into carts was cost (and even then, some producers just didn't care, and stuck ridiculous ARM7 cores into games like Megaman X3 for no useful reason at all...), I really do believe you would have seen something very much like Mario 64 for the SNES if it had lived just one more year.
(Heck, I bet that we know enough about the hardware today that someone from the demoscene could pull off a Mario-64-alike on a cycle-accurate SNES (with SuperFX) emulator. That'd be a fun competition.)