You don't appear to get it.
Yes, in *some* situations the system could defy the user's explicit
instruction to go to sleep and go on with whatever it was doing.
However:
a) This behaviour won't be consistent, and it won't be obvious to
the average user why it's not consistent.
b) Making the system work "well enough" in this context is much
harder than you can possibly imagine. You can't think of "the
system" as a monolithic entity; there are dozens of components who
each have dependencies, and you'd have to have some way of matching
the possibly degraded state of the system to the components depended
on for the current set of operations and asserting that they were
going to remain operational...
c) The user has told the system to sleep. It should sleep.
= Mike
