It's a little something that Terry Sejnowski taught me in a class almost 20
years ago.

If every cell on retina is inhibited by its neighbors, then each cell will
fire (well, photoreceptors don't have action potentials, but the idea is
the same) at some rate based on the log of its own illumination, minus the
inhibitory effects from its neighbors.  Keep in mind that if the neighbors
are illuminated at the same level, their base rate is the same as that of
the original cell.  (BTW, this does seem similar to your "multiplicative
lateral inhibition, but I haven't seen it termed like that before.  A quick
search for "lateral inhibition retina"  should yield you some gold with
respect to the basic neural substrate physiology)

If you think about this, if the inhibition is anything less than zero,  
then the receptor must be firing slower (still making believe that
photoreceptors fire, and are excited, rather than hyperpolarized, by
illumination, but it makes no difference for this derivation) than it would
for the same level of illumination shining on that receptor alone, if that
receptor were isolated from the net, and it would take a brighter light to
make it fire the same rate while in the network.  Wallah-- increased
dynamic range.  The more neurons connected together in this way, the
greater the effect.  This way, we can have exquisitely sensitive
photoreceptors, while still maintaining a useful dynamic range.  This still
isn't wide enough, though, which is why we still have two systems-- rods
and cones -- of different levels of sensitivity that work best under very
different levels of illumination.  

Note that this same network, through disinhibition, has the effect of
highlighting sudden changes in contrast--a form of edge detection, right at
early levels of retina.  

The latter is how most people think of the functionality of lateral
inhibition in retina, but the dynamic range issue is just as important.

That same year, DA Robinson showed us that exactly the same network, taking
advantage of membrane time constants, serves as a very stable neural
integrator in the brainstem (this was work by his student, Steve Cannon).  
I don't know if anyone has shown that the network works this way in retina
or not, but if the response to a light pulse were perpetuated, I wouldn't
be surprised.

Thus, lateral inhibition brings us
1- improved dynamic range
2- spatial differentiation (i.e., edge detection)
3- neural integration

It's my favorite morphology, and I try to get that across to my students.

Scott