You were doing quite well until this, Simon:
"Think about it this way, if you had a compact camera with a typically small image sensor, its photosites would be dwarfed by those of a DSLR with the same number of megapixels, but a much bigger sensor. Able to gain more information, the large DSLR photosites would be capable of turning out photos with better dynamic range, less noise and improved low light performance than its smaller-sensored sibling".
Photosite size is IRRELEVANT in the context of this discussion. At the IMAGE level, all that matters is sensor size and the technical state of the art of the sensor.
Proof? Well just look at (say) the Nikon D7000 compared to the Nikon D200: the latter has far more, and far smaller, photosites, but kicks the D200 into the middle of next week in IQ terms.
Or how about the 18 mp Canon 7D? It is not only FAR better, in IQ terms (and ESPECIALLY at four figure ISOs) than say, the old Canon 8 mp 30D; but it also handily beats older FF and APS-H bodies like the 1D Mk IIn, the photosites of which are HUGE in comparison to the 7D's.
You seem to be making the funamental error of comparing results at the "pixel" level rather than at the image level. The simple fact is that - right across the board - IQ has improved along with increased pixel count, on same-sized sensors.

An excellent and informative article. Thank you.

@ Keith Reeder: I think you have a point, but it only goes so far. You must be into conventional flash and full daylight photography to make the statements you do. I , however, am into astrophotography and the gist of this article is exactly right for that pursuit. As a bonus, my first generation Sony NEX-5 is an excellent low light camera even without its flash as long as the subject is in well balanced light to begin with. The IQ you describe applies to this camera over the 8 Meg Canon EOS20D I started with. Being APS-C format helps too with the limitations of my telescope where a full frame could suffer vignetting. Oh, and the older NEX has a lower pixel count than its successors giving more exposed area to each pixel. There is, I think, a sweet spot when it comes to trade-offs, and there is no substitute for area under low light conditions no matter how good the processing. Garbage in nets garbage out.

The other thing is Signal to noise ratio. Packing zillions of megapixels on a pinhole camera is just nuts. The signal to noise ratio is going to be so high that unless you are under ideal lighting conditions, the photo is going to come out with so much noise, it will be useless.

16 megapixels in a 2x2 mm sensor results in light hitting more than one "pixel" sensor at a time... You can increase bullshit all you want, but you can't decrease light's wavelength... 16 MP ~= 4000 x 4000 pixels; 2 mm / 4000 = 0.0005 mm / pixel or 500 nm / pixel. If there were absolutely no gaps between light sensing "pixels", you would at this point lose half the red and some of the green light since their length range goes over 500 nm, up to 740 nm.

DSLRs are great, but not for me when going on holiday. Something not mentioned in the article is the lack of zoom when using a large image sensor. With smaller sensors you can get amazing zoom levels - ideal for focusing on that interesting item in a cathedral or something in the distance. Modern high-tech compacts give you (almost) the best of all worlds. My Sony HX9V Cybershot has optical image stabilization and a 16X optical zoom in a small package. And, best of all, it uses image processing to give superb low light shots - it takes multiple short exposures and combines them to give amazing shots in low light. And you do not look like a nerd with all the gear. The ONLY problem is you lose the ability to play with focus, ie. to get the background out of focus.

Great article. I wish the cameramanufacturers would take this seriously .The pixelrace we have seen in the past, especially over the last year is just crazy. Why should a regular compact zoom camera have a sensor packed with 16 or even 18 mill pixels, while 10 or 12 mill pixels would in most cases be a much better technical solution both regarding signal/noise ratio and light sensivity as well as better pictures. Nothing but stupid marketing if you ask me.
One thing, why have you overlooked Samsung as to highend compacts and mirrorless cameras? They are definitely in the top quality brand series.

All very interesting; however, I have a Nikon D2X, D300s, D5200 and a Sony REX100. The two best shots with excellent picture detail come from my Sony REX-100 and guess what, the D5200.
The lens on test with all the Nikon’s I tested was a Nikon AF-S DX 35mm f1.8G. I did this test after taking a shot from the same position with my Sony REX-100 just to compare the shots. I was shocked when I saw the results later. I then bought the D5200 and the extra pixel in my case has made a big difference.

The same as Franc I am also a Bridge camera user. I cannot justify the expense of having a DSLR with 3 lenses, including a 800mm telephoto, which could cost an arm and a leg, and then having to lug all that equipment with you. I have shot everything from a dragonfly to a surfing competition with the same gig. (My brother-in-law has a fancy Nikon, with all the lenses, but could not shoot the dragonfly because he did not have the correct lens with him) - but I do realize that the quality of my images has been compromised by my equipment, but I am not planning on publishing any of my images. (and neiher does my brother-in-law with his fancy kit for that matter, but I had much more fun than him)
Interestingly enough Canon has actually lowered the pixel count in their latest range of bridge cameras from 14mps to 12 mps. It is the Canon PowerShot SX50 with an optical zoom of 50x. (2500mm equivalent?!!?) What would you pay for this size of lens for a DSLR? Would you get the wheelbarrow included in the price?
Thanks for a very nice article.

Whatever combination of sensor size and pixel (site) count, there is a quantum physics fundamental that decides all.
If the pitch distance between two pixel receptor sites becomes less than 1/4 wavelength of the arriving photons, then the information in both sites will be the same, as if the two together came from a single photon wave.
This is a resolution limit no amount of extra pixels can get around without increasing the lens aperture. Bigger lens focused on bigger sensor is more than simply "increasing the amount of light". The size of the available light resolution "mosaic" also gets finer. There is more information in the incoming image.
If you look at two objects on the moon through a telescope so close they can only just be seen as separate, them stop down the telescope by placing a smaller aperture mask in front of it: the image will dim, and the two points will merge into one, still perfectly focused, and any extra magnification will not help. This is why we need optics as big as Hubble and more to go after resolution.
Putting the same number of pixel (sites) on a bigger sensor, and then using optics with a big enough aperture to spread the image over the sensor is what does it.
Complicating things by using only part of the sensor, and rapidly altering the addressing to counter camera shake, or addressing only a bit in the middle to have "electronic zoom", and averaging sites to make single lower noise "new pixel" all cannot get around the fundamental physics of what is "resolution".
Resolution is the number of line pairs/mm of bright to dark transition to an agreed level. It quite hard to do. If it gets close to the pixel spacing, you get interference patterning.