Biological molecular machines are enzymes that simultaneously catalyze
two processes, one donating free energy and second accepting it. Recent
studies show that most native protein enzymes have a rich stochastic dynamics
that often manifests in fluctuating rates of the calatyzed processes and
the presence of short-term memory resulting from transient non-ergodicity.
For such dynamics, we prove the generalized fluctuation theorem predicting
a possible reduction of energy dissipation at the expence of creating some
information stored in memory. The theoretical relationships are verified in
computer simulations of random walk on a model critical complex network.
The transient utilization of memory turns out to be crucial for the movement
of protein motors and the reason for most protein machines to operate as
dimers or higher organized assemblies. Our conclusions are based on analysis
of the simulated time course of the catalyzed processes expressed by the
strings of discrete jumps at random moments of time. Since similar signals
can be registered in the experiments, all the theses of our studies are open
for experimental verification.