The Z80's at 9,000 transistors (not sure how many gates, but almost certainly a fraction of that transistor count), so even the Acorn would be heavy in comparison. Still doable, though; just takes more time.
In terms of speed, it has less to do with transistor count and more about how close together you can get the transistors. Big, hand-wired CPUs tend to be slower than small single-chip ones just from the sheer latency differences between components.
> One does not have to make an exact copy of something which made sense for arbitrary reasons, which don't necessarily apply when doing it from scratch under different circumstances.
True, and I ain't saying one does. If we're at the point where we have to hand-wire replacements, though, it helps to have at least some degree of compatibility with the thing we're replacing. There are at least some schematics out there for building 8-bit CPUs from TTL chips¹², and I'd imagine those would all be viable candidates if we have to re-bootstrap our computational power and run out of other CPUs to tide us over in the meantime.
Ideally we should be working on CollapseOS equivalents/ports for as many CPUs as possible, so that we know that no matter what we're stuck with, there's always a way to repurpose it. Just as importantly, though, we should be hoarding copies of pinout/wiring diagrams, hardware manuals, etc. to make sure we have the knowhow on the hardware side, too.
¹ http://cpuville.com/Kits/8-bit-processor-kit.html - happens to be bus-compatible with the Z80, though not ISA-compatible as far as I can tell.
² http://mycpu.thtec.org/www-mycpu-eu/index1.htm - more "modern" features like Ethernet and VGA out, so a more likely candidate for general purpose computing if we really do run out of Z80s to scavenge