The trick was to make it work without requiring an adder -- and subsequent carry propagation delay -- in the data path. Thus, "one gate delay" and simple replacement of address lines, and a simple "all zeros or all ones" check on the bounds, rather than a numeric comparison.
I ran into a related problem, I had a need to drive a whole pile of IO and only an 8 bit centronics port to work with (on the ST as well). So I ended up making a little demultiplexer using 4 bits of the 8 as address and the remaining 4 as data. The strobe was used to latch the data. Not super but it worked well enough to let the 68K drive stepper motors in real time near their maximum working frequency (30K steps / second or so).
The ST was a fascinating machine, lots of ports, relatively cheap and with all kinds of limitations that continuously drove the users to creative hacks. I loved that time. Thanks a ton for all your work.
Oh, one more for the road: we had a product that was copied quite frequently and obviously wanted to make that a bit harder. There was no budget for a dongle so in the end I looped one of the 8 bits of the port back to an input. If that wire wasn't present the software would exit with an 'out of memory' error :) Threw the copyists for a loop because all I did on not finding the wire was allocate all the remaining memory which would then cause a TOS Bomb on the next memory allocation, which could happen just about anywhere. Nasty little trick.
Edit: Ah, got it:
http://chrisacorns.computinghistory.org.uk/docs/Torch/Torch_...
Indeed, the Torch Unicorn, Unix on a 68000. Makes you wonder what they did for MMU, apparently it ran Unix System 3
There was a brown box at the time, iirc by Torch (?) that ran a version of Unix on a 68K. Does anybody know which machine that was?
The 6809 OS that was 'unix like' referenced in the comments is about OS/9.
https://en.wikipedia.org/wiki/OS-9
Which is a really neat little OS that you can run on the TRS-80 CoCo and the UK clone of it called the 'Dragon 32'.