I'm totally not clear about where this discussion is heading...
Are you talking about accelerating existing systems, or new one or ones that are prepared for this? For example after powerup the memory at $2000 in a Commodore has random content (if at all on an unexpanded vic20)

If I have an accelerator, it should have onboard (accelerated) memory. Sometimes employed in a write-through mode to e.g. set video memory on the host.

However, if I'd would create a self configuring accelerator, I'd build the accelerator as a self contained system and just map a window of address space of the host somewhere into the accelerator memory space. This way I don't care about conflicting memory locations and being able to identify the host with an arbitrary sequence of checks.

Notwithstanding that after identification the write through mode may be configured instead of a mapped memory window.

Does this make sense

_________________
Author of the GeckOS multitasking operating system, the usb65 stack, designer of the Micro-PET and many more 6502 content: http://6502.org/users/andre/