Consider

fig, axs = plt.subplots(2, 2, sharex=True, sharey=True)
for i in range(2):
    for j in range(2):
        axs[i, j].hist(np.arange(10) + 10 ** (i + j)) # widely varying scales
for ax in axs.flat:
    ax.set_xscale("log")

plt.show()

The axes end up x-relim'd to the data limits of the last axes, hiding away the other histograms (that part seems vaguely related to #3481?).

Of course, one may think "just call set_xscale("log") on one of the axes and the others will follow, right?". But switching to a single call axs[0, 0].set_xscale("log") end up as an even worse disaster; the scaling is not changed on the other axes but the tick markers are.