Hmm, interesting thread actually. I wonder if it's not a case of failing vision, but the way we process the images. The sample image is clear as day for me: fApBA
Most of these are case-sensitive. Could that be it?
I am definitely curious about what causes problems with reading these images, because I've had to use them in the past to prevent spam. If people can't read them, then it's good to know why.
I've been considering
reCAPTCHA, which provides more human-readable text, provides an embedded refresh mechanism which doesn't force you to reload the form, adds sound support and actually supports a good cause.