I have data from 2018 (pre-spherical) to compare to 2019 (spherical). In both cases the car has a wing and a splitter, but the 2019 splitter is "improved". The track is the Mosport GP (AKA CTMP GP) and the lap times are comparable at 1:34. 2018 was cooler so had more power, but 2019 corner speeds were a bit higher, giving about the same lap time.
Plotting steering angle against lap distance reveals that the spherical car requires less total steering input, despite the fact that the cornering speeds are generally higher. Lower trace is GPS speed vs distance.
Another way to plot the same data is as an x-y plot of steering wheel angle vs. lateral acceleration. Once again it is clear that less steering angle is required for the spherical car to produce the same lateral acceleration.
Actual understeer can be estimated by calculating corner radius from lateral g and speed, calculating ackermann angle from wheelbase (2.66 m) and corner radius, and comparing the ackermann to the actual steered angle at the tire by dividing the logged steering wheel input by the steering ratio (16.36 near center). The method is mentioned in Bob Knox's "A Practical Guide to Race Car Data Analysis" if you want to chase down the details and the limitations of the method. Whatever its limitations, it is probably good for comparing two setups of the same car, as we have here. This method shows a reduction of understeer mid corner of up to 1°.
My hypothesis is that going spherical (and solid RSFBs) reduces compliance steer effects and loss of camber. This reduction provides greater total grip, reduced understeer, and thus less total steering input being required.