However, the datalim seems so implicitly include zero.
> For example, if min(y_values) is 3, I still see the point
> of origin. How can I avoid this behaviour?
We're just doing what we're told here 
Your first plot call is
y_values = total_number_values * [0.0]
line, = plot(x_values, y_values)
So zero is *explicitly* included. The datalim remember the history of
objects have have been added. If you want them to ignore their
history, call
ax.dataLim.update_numerix(x_values, y_values, True)
The third argument if True will cause the datalim to update it's
bounds ignoring history. If False, it will update the bounds
including it's past history.
Note that in the original example you posted
for i in range(2*cycles):
voltages = number_of_values * [i % 2 + 3]
y_values[i * number_of_values : (i+1) * number_of_values] = \
voltages
line.set_ydata(y_values)
print min(y_values), max(y_values)
ax.dataLim.update_numerix(x_values, y_values, True)
min(y_values) == 0
JDH