This makes sense only if you’re rolling your own silicon, and it depends on the application. I could see this as an alternative to a scalar DSP core or to provide more flexible high speed interfaces. The unspoken cost is going to be adding another proprietary set of tooling in the BSP for your chip.
25 years ago I worked at a university computer lab that was Windows-heavy because Dell wouldn’t stop donating PCs. However we didn’t have enough UNIX workstations as we had to pay for Sun / HP / IBM out of pocket. Converting them to Linux workstations would be nice because the Dells had more grunt than the aging RISC workstations.
I proposed to switch a few desks worth to Debian and was given the go-ahead. After a few days learning how to preseed an installation image and getting a PXE server going I had 8 machines running CDE just like the AIX and HP/UX boxes. Users that didn’t need one of the commercial engineering applications tied to one OS or another didn’t notice any difference between the free (now as in both speech and beer) Dells and the proprietary workstations.
A couple of months after we got the pilot rolling, the university’s IT director came to check it out and told me we’re on the “lunatic fringe” for deploying an OS developed by volunteers, but otherwise offered approval as long as we could maintain security and availability.
Now every student in our local school district gets issued a Chromebook running Linux under the hood. Who’s the lunatic now?
I never used Itanium, but I’m guessing that the Alpha workstations also ran x86 code faster than the Itaniums. fx!32 was one of DEC’s marvels that they completely forgot to market.