The OP doesn't say that a single retinal cell transmits 10 million bits a second, but that the whole eye does. On top of that, while discussion of collition detection is pointless, thinking about the information a neuronal population can encode does have some merits. Although it's relatively pointless (at least now) to compare the eye to an ethernet, it has uses in comparing different neural populations.

The problem is that getting bitrates for neuronal populations is more of an art that a science. The sum total of information passed on by a neuron can not be computed simpley by it's spiking rate. Large numbers of parameters alter the actual chemical I/O relationship of a neuron. Resting membrane potential before spiking, whether it shows short term facilitation/depression etc...