Your test case shouldn't be as fragile as you suggest. Interrupts arriving "between" EI and PCON aren't easy to schedule and test, but anything that happened during the entire period since you DI is going to be waiting to pounce when you EI.

I would propose (as a theory to be verified) that if interrupts that arrive between DI and EI don't get processed before you sleep, that interrupts arriving later than that shouldn't either.

regards,
-Mike