I'm not sure what happened. Doing the math, the S3 image is 8MP, but the 640 image is only 2.1MP. Please check for yourself:
Camera Resolution/MP Calculator
Is it the way you uploaded the pics? Did you share the 640 pic to yourself in email or some other way that may have auto-reduced size?
What camera App did you use to capture the image with Lumia 640? Is it possible you have it set to use a reduced capture size?
The size disparity makes it hard to give a good side by side comparison. It made me wonder when I looked at the images full size in my browser as to why the 640 image appeared so much smaller at 100%.
Let's take a look at what the cameras did with what they have.
The Lumia 640 has an aperture (F-stop) of f/2.2 which allows it to gather light better than the Galaxy S3 at f/2.6. This is reflected in the Lumia using a higher ISO and faster shutter. Flash was off on both so it did not factor. The Galaxy S3 used a shutter speed of only 1/30 of a second (twice as slow as Lumia) and that is probably the major factor for lack of focus. In general, the higher F-stop # on the S3 should give it a greater depth of field (more focus as you move nearer or farther than the subject), but it was defeated by that slow shutter which translates "wobble" without any stabilization. I prefer the cleaner depth of field that Lumia 640 captured.
I ran into another anomaly:
The Lumia identified the light source as Tungsten. That means old style incandescent bulbs. They definitely have a yellow cast to the light.
But apparently you adjusted the White Balance on both cameras. I have no idea what you did for an individual setting or even if you set them both in the same direction.
The Galaxy S3 interpreted the shot in a much colder temperature. The colors are decidedly slanted towards the blue end of the spectrum. It is most noticeable on the ceiling structure and the gi uniforms of the participants. Maybe you were adjusting White Balance to achieve an effect? Not sure what that was until you report back.
Is it possible to get two fully auto samples from similar conditions and both at full size? Makes it hard to judge otherwise.