Biological neurons have a trade-off between high information transfer (high firing rate) and energy conservation (low firing rate). One would suspect that the maximization of this function has a single solution, and that the mean firing rate and variability should be similar across all neurons. Yet this is far from the case. Is there a principled computational explanation for this variability?
I am not sure whether the three assumptions on which your question is based are really valid.
(1) Why should a high transformation transfer be linked to a high firing rate? Depending on the role of a single neuron within a group, not firing might carry as much as information as firing.
(2) Energy conservation might not be linked to the behavior of a single neuron, but rather to the organism as a whole. For instance, if the constant firing of a neuron stops your legs from constant (dis-functional) shaking, the energy "wasted" on the firing neuron can lower the energy consumption of the system as a whole.
(3) Although not explicitly stated, your question might suggest that the mean firing rate of a given neuron is relatively constant over time. This is not necessarily the case, as the firing characteristics of neurons are "tuned" over time, depending on its activity. This process is called homeostatic synaptic plasticity and has received much attention recently.
For a review see: Turrigiano GG. (2008) The self-tuning neuron: synaptic scaling of excitatory synapses. Cell. 422-35.