The SFML RenderWindow is already double buffered. The method draw(...) draws to the backbuffer and display() swaps the buffers. So there is no need for a RenderTexture in this case.
The texture's copy constructor loads the texture from the gpu in an image and creates a new texture from that image. Both done per pixel and as cpu operation which is definitely slow.