I think it may be the measurement technique itself that's causing the odd results. The problem is that the anemometer has an effect on the flow. It's clear from just looking at it that the anemometer housing is an obstruction and reduces the effective area of the duct.
Most of the time this type of measurement involves a duct that is larger than the anemometer and the main effect is to cause the velocity reading to be higher than without the anemometer in place. This effect is documented in a
US Bureau of Mines report.
In your situation I think something else is happening. Not only is there an area reduction but also a friction loss increase that's larger than you would expect based on area alone. That is, the cross section perpendicular to flow is not an open circle and so has more resistance than a circle of the same area. In addition, even though they are spinning, the vanes also cause a pressure loss.
It's possible that with a sensor that's the same size as the duct it could be the highest loss part of the path. In other words, it's a choke point that keeps your other changes from showing much difference.