The radalts I have worked with are comfortable with a composite attitude angle of around 45°. This would roughly correspond to the half cone angle. I have to say that this is a general observation and is very much dependant on individual units, and their antenna designs and placements.

I think that is generally correct, in level flight over a "smooth" reflective surface the ground speed would have little to no effect.

Errors come from a variety of sources, modern radalts are very good at coping with scatter and multi-path effects. They are also largely doppler immune. By design they look for the leading edge of a valid return signal to try and give the closest point and not the slant angle. Therefore if the ground is closer than the slant range to the building then it will still track the ground. There are also filters that prevent sudden large changes of altitude, such as when passing over another aircraft. Accuracy is generally given as a percentage so therefore the lower you are the more accurate the reading. Although there is a lower cutoff due to the limit of technology that gives an absolute measurement accuracy usually around ±3 ft. Because a descent rate is seen directly by the altimeter excessive descent rates can degrade the accuracy as well.

Thanks for that. It seems I need to refine the angle question a bit. The footprint may not be circular. It is the roll angle that I was wondering about. Do you know the angle from vertical of the outer edge of the beam, side-on to the direction of travel?

Unfortunately I don't, sorry.
For an approximation of the radalts I know, I'd assume it is circular and at 45°. The tech specs that I have don't specify individual angles, just an overall angle limit. But looking at the TSO it sepcifies a test limit in roll that is greater than in pitch, which makes sense in operation.