The Pentium Pro, released in late 1995, was a 3-way design clocking up to 200 MHz. It was surprisingly competitive. It had all the modern requirements: out of order execution, out of order memory pipeline, on board L1 and L2 caches. It just had less resources in each dimension. Intel caught up to 500 MHz in 1999, with the PIII. That was still a 3-day design. Intel’s first four way design was the Core 2 in 2006. That could hold 96 instructions in flight. Intel didn’t release a CPU with two load store units until 2011 (Sandy Bridge could do two loads per cycle) or possibly 2019 (Sunny Cove was the first that could do two stores per cycle).
Obviously, Intel caught up based on a combination of features long before then. I’d guess around early 2000s, when clock speed hit 1 GHz+. Clock for clock, the Pentium Pro wasn’t actually that much slower than Alpha on integer code. It took much longer for intel to be competitive on floating point code.
(By the way, it blows my mind that we went from 200 MHz to 1 GHz in about 4 years. Meanwhile, it’s been 15 years since my first 2 GHz+ MacBook, and a new MacBook still has a base clock around that range. Maybe it’ll turbo boost to 4 GHz for 10 seconds before it exceeds its thermal envelope.)