Itanium cpu was built by HP and Intel cooperation for long instruction execution. Optimized for 64 bit which leaves gapping holes on each clock cycle when 32 bit software is used. Your guess of half the speed with 32 bit programs would be about right.

Digital equipment now HP, started with MIPS technology to build the alpha cpu with two 32 bit cpu appended for 64 bit execution of 32 bit software. So, no difference in speed for either long instructions or shorter instructions. Original Athlon uses the same pinout as alpha and was licensed by HP.

IBM had many years ago designed and build the IBM370 cards for original PC. I owned two of the 64 bit computer cards for one 370 computer. The PC became the terminal running 370 64bit computer on ISA slots. The two cpu used were Motorola 68040(subsequently they coorporated on PowerPC cpu). So the technology was well known in the 1980s. Compared with the original 370 designed by Gene Amdaul; there was one instruction missing for the 370 cards.

The death os Iteniumn was premature, but 64 bit software was not available, even today. The reason is the invention of branching prediction, done by Intel on their Pentium cpus. 4 instructions done on the same clock cycle by prediction, made long instructions not necessary in future computations.

So be it. The $64,000 question was preempted.