Hi,
This is a fine example of an interaction which is not entirely 1:1. You matched the buffer well and the response is fine. However, as you noted is the fitted curve in both situations (RI=0 or RI=fitted) not following your data.
You can try several things to get the interaction 1:1.
1) Immobilize less, because the response is high enough. Lower responses tend to have less non-ideal behaviour.
2) Check ligand and analyte for purity (gel). If there are extra bands or breakdown products, try to get rid of them.
3) Add some extra P20 (up to 0.01%) or NaCl (up to 250 mM) to suppress non-specific interactions.
Arnoud