Why are these file sizes not normally distributed? I have saved 10,000 webcam images and tallied their lengths.  The lighting conditions were constant throughout the recording time. The probability distribution is shown here, with my best efforts at fitting a normal curve to it...

The files are created via multiple complicated processes of quantum mechanics, electronic theory, temperature and of course the JPEG encoding algorithm.  You can see that there is a skew to the distribution. This is apparent even without the curve fit.
Q. After 10,000 samples, why is the distribution not 'more' normal in accordance with the central limit theorem?
 A: CLT applies to the mean of samples from a distribution. If you bootstrap the mean of your data you'll find that the distribution of the mean of your data is normal. The file sizes of the images are neither generated under these conditions, nor do they have the assumptions in which the CLT apply, so it's not expected that the distribution be normal.
A: Please take a look of the definition for Central-Limit-Theorem:

This statement tells us that you only get a normal distribution for the sample means given enough sample size. There is no assumption to the underlying distribution that it needs to be normal.
A: The standard central limit theorem holds under some conditions, one of which is independance of observed events. Were these captured images in such a way that parameters of the images had no correlations between them ?
It's kind of reassuring to think that if you were filming a periodic phenomenon (say, the position of the sun at the same time every day), you wouldn't end up with a normal distribution in parameters of the images.
There are some other theorems showing that convergence to the normal distribution still occurs when events are not totally uncorrelated. These theorems rely on complex hypotheses, some of them stating that observed events have to be "not too much correlated".
