He did say that and he's wrong. x86 isn't particularly dense and RISC-V compressed is denser. However the insane encoding is burning real power, takes up real area, and is a real limiter for decoding. Dear Linus has an irrational love for x86.
He wrote those comments in 2003. The instruction decode logic takes up less space with each new chip and process shrink. These days I don't think it really matters much.
Oh, I don't think the link was originally there, or I missed it. Yes, of course: that claim is like the first shot across the bow in every CISC vs. RISC debate.
Linus says to ignore the design mistakes like segmentation.
Then he says:
"the baroque instruction encoding on the x86 is actually a _good_ thing: it's a rather dense encoding, which means that you win on icache."