By way of example, recall that in the first month after Intel started shipping its first 64-bit x86 processor (a NetBurst Xeon whose code name I have forgotten), that processor outsold the combined annual production of G5, Athlon64, AND Opteron.
AMD had a lead in many ways, but wasn't even in the ballpark as far as fabrication capacity.
"The efforts that fail almost always assume that the new system will be 100% capable on day 1 and the switchover will be effortless and complete instantaneously (rather than costly, time consuming, and incremental)."
Agreed -- and it's all the more reason that Intel should have relied exclusively on the binary translation technology that they inherited along with the Alpha development team, technology, and intellectual property.
Itanium is a special case of the classic problem of trying to rebuild from scratch. If you do it right it can be a huge win, if you do it wrong you can kill the company. Generally speaking most times that such efforts fail it's due to ignoring the need to take the new architecture through enough iterations to gain maturity and parity with the old architecture, and ignoring the extensive costs of having 2 parallel architectures in play at the same time and the costs associated with transitioning from one to the other. The efforts that fail almost always assume that the new system will be 100% capable on day 1 and the switchover will be effortless and complete instantaneously (rather than costly, time consuming, and incremental).