wiener filter basically reduces the mean square error between the corrupted and uncorrupted images. 

whereas the NLM filter finds only the similarity between the pixels in the image with the target pixel and gives the weights accordingly. here there is no consideration of reducing the error. 

you might be using PSNR as the barometer for measuring the performance. obviously it shows that Wiener filter performs well over NLM filter. 

wiener filter has its own advantages. similarly NLM filter has its own too. 

it all depends on with what you are measuring their performances. 

thank you.

if you use the standard versions of those methods, both of them are not suitable for attenuating the rician noise which is the type of noise that degrade MR images. instead, you can use TV based denoising methods.