*IF* I remember correctly, the 68010 had some built in support for virtual memory, no memory protection though.
The 68010 could probably use an external MMU but not in the way the Motorola ones worked. I *think* the MMU would generated interrupts through the standard lines on the CPU and the interrupt handlers had to read from the chip through memory mapped I/O rather than special MMU instructions and registers.
FYI, the speedup came from using the CPU prefectch memory to allow one (two???) instruction loops to take place without reloading instructions. Sort of an instruction cache. That's why you won't see a noticable difference in benchmarks. It's the same clock speed and most loops are longer than that.
I did write some stuff that took advantage of it and the difference was noticable. It sped up some color fade in/out routines and I was able to do color cycling in an interupt without garbling the Amiga speech. On a 68000 we had to turn off the color cycle when using speech. As it turned out... the other programmers didn't want to check what CPU they were on so they always turned off the color cycle to use the speech.