Lol, simply untrue. We were (and still are) locked in x86 architecture by applications. Business and consumers demand backwards compatibility. Without AMD64, we simply would have been stuck in 32-bit land longer until Intel came up with something similar.
No we wouldn't. Because if Intel is the only supplier of x86, Intel decides not to come up with 'something similar', meaning that people will HAVE to upgrade to IA64. The same happened with Apple multiple times. The Mac went from 68000 to PowerPC, and from PowerPC to x86, simply because 68000 and PPC ceased to exist.
When x86 ceases to be a moving target, IA64 would exceed it with x86-emulation performance eventually. But people would also be fully aware of the fact that native applications would be faster than x86-emulation, so a demand for native IA64 applications would arrive. Moreso than AMD64 has done so far.
IA64 was not superior, unless you think expensive, slow, and difficult to optimize for is somehow better.
Firstly, whether IA64 itself was superior or not wasn't the point.
The point was, IA64 would cut the binary ties to x86, opening the doors for future superior architectures.
Secondly, difficult to optimize for applies to x86 more than IA64. Compilers take care of that for mortal developers.
At least for IA64 the performance of the assembly code was very explicit, much like with RISC CPUs. With x86 there are tons of pitfalls 'under the hood', that you can't tell by just looking at the x86 code, because of all sorts of decoding rules, register renaming, aliasing, synchronization issues etc at the execution backend level, which you cannot program directly. Most people who *think* they know how to write x86 asm, can't hold a candle to a modern compiler because they have no clue about this.
Lastly, IA64 wasn't slow at all. It scored extremely well in various industry benchmarks. Especially its floating point performance was 'best-in-class', well out of reach of x86. This made IA64 quite popular in HPC situations.
IA64 cores are also smaller than x86 ones, so it would have scaled better in the multicore-era. More cores in the same die-space (the IA64 dies were mainly large because of their extreme amounts of cache, given their target market).