If you just want a box with a gradient in it then performance isn't really an issue. You should go with whatever is simplest to implement for your particular requirements.
Adding a CAGradientLayer means you don't have to create a view subclass, you can just add the layer to an existing view. The setup is also slightly easier, since you don't need to worry about frame sizes or any c-style core graphics functions. You can also add rounded corners, shadows etc without too much effort.
However, if you want more than one gradient view, a subclass might be a good idea, so you can just instatiate new ones.
So, unfortunately, there isn't a clear cut answer to your question. Neither is definitively better. If you are concerned with performance, implement both and test using instruments.