As a degenerate case, can anyone design a simple 'meta machine' or constraint-set in the instructions which imposes specific costs on the interrupt decision logic? Can one code to this? Is this useful or just entertaining?

It might (for instance) be analogous to designing solutions which fit in L1 cache, for the outcome in time to execute.

Or it might affect the side channel consequences of some forms of crypto.

(or am I missing the obvious "this is in fact, how it works" answer. I am sure I recall in simpler times reading that writing interrupt handlers demanded some consideration of what was fixed or mutable in the throw)