How it connects is irrelevant.
It is if you are trying to insist that it makes the hardware "frankenstein". It's about the most relevant factor there is.
The PPC is not a 68K.
It isn't? I was robbed! :lol:
The CPU IS the brain in a computer. The PPC is a foreign brain, inside a foreign body, connected to, communicating with and powered by original pathways, analogous to the peripheral blood supply and peripheral nervous system. Frankenstein all over.
You already said, pre-OS4 (not to mention MOS and ppc linux), it is implemented as a co-processor, not
the CPU and to be fair, it's not a bad analogy. 68K code gets the PPC to do some processing for it. That processing can be just a couple of functions in an-otherwise entirely 68K application. Or, it can be pretty much the entire application, but it is still launched by the 68K and control is returned when it exits (not to mention any time it does a system call).
I notice you studiously avoided commenting on the SCSI script processor. Lots of accelerator cards have those. They are given a list of instructions by the 68K and they go away, do their work and return. Viewed implementation terms it's clear there's not a lot of difference between how they operate and how the PPC does in a 3.x environment, other than the fact the PPC is capable of doing rather more varied things than talking to SCSI devices and transferring data to and from memory. Unlike the very fixed-purpose SCSI controller, it functions, essentially, as a Turing-complete general purpose co-processor.