Performance Check of OpenGL interoperability (textures)

I am using CUDA/OpenGL interoperability and I would like to check that the runtimes I am recording match the expected performance. I am transferring OpenGL textures to CUDA with the following signatures:

typedef texture<uchar, 2, cudaReadModeElementType> GrayscaleTexture2D;
typedef texture<float4, 2, cudaReadModeElementType> Float4Texture2D;

These textures are the one that assemble my custom (offscreen) OpenGL framebuffer and are all of QVGA size: 320*240. The framebuffer is bound to CUDA by functions like these:

checkCudaErrors(cudaGraphicsMapResources(1, &res.resouce, 0));
checkCudaErrors(cudaGraphicsSubResourceGetMappedArray(&res.array, res.resouce, 0, 0));

These two resource mapping functions take a median time of 1.25ms.
Is this a reasonable expected execution time for an NVIDIA GeForce GT 650M?

What I don’t understand is that those resources are already on the GPU and in read-only mode.
Shouldn’t the times should be in the range of microseconds, instead of milliseconds?

Thanks for your feedback!