We had this discussion on here before and someone much cleverer than meade the point that as well as size of sample, what is key is how the sample is selected. So a truly random sample of 800 people could in some circumstances be a v good indicator of a much larger population. That is why randomised nationwide polls can have a bit more value - the goonhammer web poll is more akin to a poll that is put on a political parties own web page in terms of randomisation - and that's why it's right to be very sceptical of the data.
That said, that doesn't mean the data isn't useful or interesting and I certainly don't mean to imply that goonhammer act in bad faith like a political party would do!