No, they do not need to be equidistant or equiradial. This is true for 4-wheel or 3-wheel configurations (and would also be true for 5- 6- or 8-wheel configurations if you wanted to make one).
When the robot is translating in a straight line, the math involved to determine the rotational velocity of each drive wheel is independent of the distance from the center of motion to the wheel (it is dependent only on the angle between the CM-wheel line and the desired direction of motion). So the algorithm used to determine the drive parameter for a configuration with wheels that are not equidistant will be identical to one that is equidistant.
When the robot is rotating about a point, the math involved to determine the rotational velocity of each drive wheel is dependent on the distance from the center of motion to the wheel. For a system with all wheels equidistant from the center of motion, the algorithm to determine the desired rotational velocity would be executed once, and the results applied to all wheels. For a system with each wheel at a different distance, the algorithm to determine the desired rotational velocity is the same, but the offset parameter (the distance from the center of motion to the point of contact between the wheel and the ground) is different. The algorithm is executed once for each wheel, and the results applied to that wheel. In other words, the math is all the same, but you just do it once for each wheel instead of once for the whole robot.
The cambered holonomic drive system that Team 116 developed this year used four drive wheels that were located at different radial positions and different distances from the CG. The system handled this configuration without issue.
If you think carefully about how this works, you will find that you can start to have some real fun with holonomic systems. Unlike the center of gravity (which is fixed for a given configuration of the physical components), the center of motion is a virtual point that can be relocated. You can change the location of the center of motion just by modifying the offset distance between each wheel and the center of motion, and completing the associated calculations. Then apply the algorithm used to rotate about a point, modified with the relocated center of motion, to determine the drive parameters. In this way the robot does not always have to rotate around the center of mass or center of volume. Instead, it can do things like rotate around a corner of the robot, or around one wheel ("twirling on it's toes"). Place the center of rotation well outside the robot volume and orthogonal to the direction of motion, and it will drive in an arc. Place the center of motion outside of the robot and directly in line with the direction of motion, and it will linearly translate. In fact, all desired motions can be simplified into a series of rotations about a set of points.
-dave