An important factor is whether it can be branch predicted. If the virtual call nearly always goes to the same method implementation (e.g., always an OpenGL ren

An important factor is whether it can be branch predicted. If the virtual call nearly always goes to the same method implementation (e.g., always an OpenGL renderer or always a DirectX renderer), it can actually be pretty cheap. They get expensive mainly when the call is constantly jumping to implementations for different object types.