Is there a way to disable the instruction prefetcher on actual x86 (ia32) cpu? This (maybe?) will force the cpu to fetch instructions like Z80, so actual instruction execution can be traced physically on the cpu buses, and instructions can be easily profiled for documentation/emulation purposes.

I personally doubt it somewhat, as the ISA is so f....ed-up (with instructions being 1-15 bytes) and different x86 cpus have 8-64 bits data buses, so such execution will be traumatizingly slow, but for debug/analysis purposes seems pragmatic. I know about the branch trace messages (rumors have it that even 386 supports it via undocumented DR7 bits), but they're still implicit way of observing/measuring execution.