I saw Alan Kay speak once, I think at lambda jam 2013. He made a comment that he’s still waiting for mainstream CPUs to allow us to fit an entire interpreter into on-core cache.
Eh? The LuaJIT and K interpreters definitely fit into L1 instruction cache. (And have since well before 2013 probably.) Honestly I would be kind of surprised if most interpreters for reasonably small languages don't fit into icache. Sure Python's not going to any time soon, but most interpreters are really not very large.
You could get Windows NT 4 running in a processor cache these days.
Forth interpreters probably fit into L1 on modern processors.
And we've had 256KB L2 since the Pentium Pro. That's big enough for a Lua interpreter.