<i>Murphy states that that routine will grow till the time is too long.</i>

That's actually Parkinson, isn't it? ;-)

The same problem can, of course, also afflict Interrupt Service Routines.

I think the biggest "problem" with interrupts is given by the name - they can <b>interrupt</b> anything at any time. But, of course, this is also their strength!
A spurious interrupt can cause all sorts of weird effects and be extremely hard to debug but, with proper design & precautions, interrupts are definitely the way to go.

People who like to be able to mathematically "prove" their software find Interrupts very difficult.
(Hence I think it was the Viper processor which had <i><b>no</b></i> interrupts?)

Another disadvantage of polling is that you "waste" a lot of time in all the polls which just find that there's nothing to do!