The dark red bar at the top is the result of a subset of sites that have a strong, continuous (non-localized) signal over the full 3000bp range. This signal is skewing the scale so that the remainder of sites are showing only faintly.

It might be worth trying to identify the sites showing this non-localized, saturated signal. I suspect an MA plot would show a cluster of sites with very high concentrations (X-axis). You can retrieve the binding matrix and look for the sites with the highest counts, then see what they look like in a browser. You may want to concentrate on the fourth sample which has the highest read concentrations. You may also want to check the read counts normalized and non-nomalized to make sure this isn't an artefact of normalization.