From my own experience, if you're using angles much at all you're doing it wrong. First, the cube angle should be reduced to reasonable range quite often. Better in 3d is to use a different representation for orientation - a matrix or a quaternion are common and both should be "normalized" regularly. These also have the advantage of eliminating trig functions from the code.
In my day job doing motor control, I track shaft angle using a 16 bit fixed-point integer. It wraps around perfectly and I've got an awesome sin(x) function that takes -32768 to 32767 as input (equivalent to -pi to +pi) and returns a fixed point value in a few operations.
It's all a matter of using a representation that make whatever you're doing easy to compute.
In my day job doing motor control, I track shaft angle using a 16 bit fixed-point integer. It wraps around perfectly and I've got an awesome sin(x) function that takes -32768 to 32767 as input (equivalent to -pi to +pi) and returns a fixed point value in a few operations.
It's all a matter of using a representation that make whatever you're doing easy to compute.