It was not awesome to keep adjusting out code to be more amenable to the VLIW-like needs of the processor (Intel’s compiler actually generated very good messages as to “why we didn’t do a good job with your code”, you just often couldn’t do anything with that info).
I still remain amused to guess what kinds of programs were included in the architecture simulations that led them to want so many integer units (12?) and so few floating point ones.