To cut a long story short,
acpi=ht isn’t the magic solution. Although much more stable than before, stable enough to get X windows up and running, it still fell foul of the same runaway
kacpid problem eventually. It had to be another part of ACPI causing the problems. A quick look back through the dmesg output showed only very minimal portions of the ACPI code enabled. Immediately obvious was the output describing the LAPIC and IOAPIC setup. Both of these shared one very important feature: They could be individually disabled from the kernel command-line for testing.
noapic appended to
acpi=ht appeared again to be successful. After eight hours of constant use under heavy load (recompiling more optimised packages) the system was stable. Could this be the solution? Was the IOAPIC to blame?
A final reboot would confirm my suspicion: Remove
acpi=ht entirely and leave only
noapic on the kernel command-line. Another twenty hours later and this is being typed on what is seemingly a stable system.
Hopefully this will save somebody some time.