How does the SUBLEQ architecture support SMP? Does it have any locking or equivalent primitives?
Due to the reliance on self modifying code I doubt normal SMP is going to work at all (unless you like the idea of "twice as many CPUs makes it half as fast"). Instead, you'd have to shift to some kind of "each CPU has it's own separate memory" model just to get a benefit from multiple CPUs, and in that case you don't really need locking or equivalent primitives (e.g. actor model).
Fortunately, most CPUs support paging so it'd be easy to emulate "each CPU has it's own separate memory", and SUBLEQ can avoid the need to have different copies of each executable (where each copy is compiled for each CPUs different physical address range). This is why Geri loves paging so much!
I'll admit it's hard to pick out the worst part of this "tutorial," but... WTF is this??? x86 instructions aren't documented and aren't deterministic? Really? Dude, you not understanding how the CPU works is not the same as it being random and undocumented.
To be fair, 80x86 isn't deterministic - instructions take a different amount of time under various conditions (and that's without considering SMM). Essentially, it breaks the "all instructions are always slow" assumption by making some instructions faster sometimes.