That was more or less intentional. The 68K interpreter was quite fast and 68K asm is smaller in memory than PPC asm, so converting all the cold code over would've made the system slower.
I'd like to see the same idea used today. One I can think of is C++ exception unwinding which uses a DWARF interpreter instead of generating tons of code to do it.
You’re looking through rose colored glasses. My PowerMac 6100/60 ran 68K apps slower than my LCII with a 68030-40Mhz accelerator. The original 1st gen 68K emulator was really bad.
Connectix became famous for shipping SpeedDoubler, a much better emulator. But my 6100/60 could still barely keep up with the 68030-40Mhz.
I'd like to see the same idea used today. One I can think of is C++ exception unwinding which uses a DWARF interpreter instead of generating tons of code to do it.