The essence of OCR is the tollerences and patterns used for recognition. These vary from engine to engine. The software has to firstly "find" your number plate in the image. So placing your numberl plate off centre will challenge the engine under load. It also has to differentiate any other letters on the vehicle from the "Number Plate" characters. There is a limit on the tollerence of angles horizontally - some software use deskew techniques to do so but are also places extra strain on the engine and costs machine time - very taxing under high volume and load conditions. Spacing between characters are also important. For eample if your's was "G P" versus "GP" signifies two "words" versus one. The clarity of the letters - i.e. faded letters also cause strain and may fail pattern matching.
So it would be useful to find out what OCR engines are being used and see what the tolerences are and maybe there are some loopholes. 
I was wondering why, if these cameras are so good, it would be necessary for e-tags at all.
If, for arguments sake you had an account with Sanral and your number plate is listed there, why would they need the E-tag to make the connection and not just use the number plate recognition software.
It seems from the above, that it is still a lot more difficult to do it via the camera?