One interesting thing that may help the final output quality is preserving the detail layer of the original image, and then applying that to the output image. Here's my quick attempt at it: http://dev.jjcm.org/tonetransfer/
I basically just used a technique called frequency separation - it's extremely quick with the most computational part being a gaussian blur, and it allows you to separate detail from tone into two separate layers. From there I just took the detail layer of the original image and applied it to the tone layer of the output image.