I believe IBM mainframes have traditionally used something like that: binary code is shipped for a general mainframe architecture, and on first execution is spe

I believe IBM mainframes have traditionally used something like that: binary code is shipped for a general mainframe architecture, and on first execution is specialized to the hardware / performance characteristics of the particular model within that architecture that you're running. Also allows for transparent upgrades, since if you migrate to a new model, the binary will re-specialize itself on the next execution, (ideally) taking advantage of whatever fancy new hardware you bought.