This is also a note on Ben Crowell's answer (it started as a comment but I think it makes a point which isn't in the other two answers).
As Ben says, the total number of flips you observe is finite, say $n$. But the intervals you observe between the flips are not constant. In particular, if you adjust the timing mechanism on the torch suitably before dropping it, you can arrange things so the time you have to wait (your proper time) before you see the $n$th (last) flip is arbitrarily large.
As A.V.S. also points out, the total amount of energy emitted by the torch before it crosses the horizon is finite. A consequence of this is that it becomes increasingly hard to observe the flips, and in particular the last flip: the light you see becomes very faint and very red-shifted. Especially if you assume that the torch emits a stream of photons, you quite quickly (quickly in your proper time) reach the point where the probability of you being able to observe any more photons from the torch is vanishingly low. At that point there can be no real observational difference whether the torch has crossed the horizon or not: there is always a chance that you might detect another photon, but that chance is vanishingly small.
In fact see A.V.S.'s comment below: the photons pretty rapidly drop below the background Hawking radiation from the BH in fact, so it really is the case that the infalling object becomes undetectable, even in theory.