Tampering is hard enough to spot if you have full access to the server, from the outside it would be almost impossible.
But there's a wider issue than tampering.
Even if you could 100% guarantee that the poll hadn't been tampered with it still wouldn't be valid.
Think of it this way - put the same poll about Linux vs. Windows (say "Which is better, Linux or Windows") on Slashdot and on Microsoft.com. Do you think you'd get the same results? Of course not.
The reason are (and this pretty much applies to all on-line polls other than those done by polling organisations on an invite basis):
1) The group you're polling is self selected. Not everyone will have chosen to vote therefore the opinion of those in the middle ground are either missed or under representated as those with extreme opinions are more likely to vote
2) It will have an intrinsic bias based on the site doing the poll (as in the above example). A site might appeal to those with a particular view or just to a particular demographic who are not representative of the world as a whole. Which leads to...
3) The results won't be weighted to make it representative of the whole group in question. So if your website is popular among men but you're interested in opinion of the population as a whole you need to balance that so the votes of men represent only 50% of the whole. How is that achieved in an on-line poll? Simply it isn't.
4) (Thanks to @TFM) The question can be biased. For example questions with a negative encouragement: "Would you consider Linux even if it's complicated to install?"
Essentially on-line polls will tell you nothing you can rely on. If they are right it's as likely to be luck as anything else.