I don't have easy answers to this but I have observed a few things that I can share.
One is that, counterintuitively, it is essential to STOP telling the child not to guess!
I'll bet I have company out there that feel a compulsion to remind the child not to guess, to compliment him or her on correctly decoding a word with a comment like, "See how well you can read when you don't guess?" And so on. I've even made little signs with that red circle logo with the slash through it, and the word GUESSING in the middle. Now with most kids we do see improvement over time, as their decoding improves. The Reading Recovery kids are the hardest to convince.
But last year there was one who was (and remains) temperamentally, and maybe neurologically, an inveterate guesser. He does have some neurologically based issues, impulsivity being one (it affects other areas, not just reading). He learned his decoding skills pretty solidly, with lots of practice. But he was still guessing up a storm. I tried an experiment, the kind of thing you do in graduate school, and I should do more often. While another person was working with him, I counted (for a ten-minute period) the number of times s/he mentioned guessing -- to remind him not to do it, to pump him up -- "See if you can read these without guessing," complimenting him on decoding well "without guessing," and so forth. Then I also counted the number of times the student obviously impulsively read something that wasn't there -- his usual "guessing."
The results were interesting. There was a close relationship to mention of guessing and incidence of the same -- the more he was reminded (all this in a very positive and upbeat way!) not to guess, or praised for not doing it, the more it occurred. The lines on the chart were practically parallel -- the relationship was obvious.
Then we tried, for a couple of days, simply not mentioning guessing at all, either positively or negatively. If the student guessed, he was required to correct the error, but nothing was said. His guessing rate dropped, about 20%. Finally, we developed a menu of things to say that did not refer to guessing, but strengthened the behaviours we wanted to see. We came up with prompts like these:
Remember to say the sounds in the new words.
These words are hard but you can figure them out.
You figured it out -- good job!
Wow, you remembered that new sound (morpheme/word)!
And so on.
Guess what happened?
The "guessing" rate dropped a lot more!
I don't think we will ever overcome this child's impulsivity completely, but he is gradually developing a strong secondary strategy which will -- we hope -- be in use most of the time.
With older students (your KS3 age) I second Debbie's suggestion that the materials are important. So are setting goals that are achievable only by decoding, not guessing -- such as rate and accuracy requirements that can not be met if the student is guessing. I am sure you must have UK equivalents, but I found the Corrective Reading Program materials to fill this bill nicely. They are specially written to trip up the guessers, and the only way the student can meet his required "checkout" every day with 98% accuracy at 90-150+ WCPM (depending on what level he is at) is by decoding accurately and well. He is reinforced for doing it right and gets no feedback (even acknowledgement) for guessing. Every error is corrected with no comment and the student has to re-read the word and then the sentence. Most of my 11-14 year old guessers DO stop this habit within a few months of this routine at most. Many become very good readers, even outstanding ones.
I think we may often reinforce guessing without meaning to, just by paying attention to it. Now I hope no one will take offence, but I learned a valuable teaching principle from dog training (I realize kids are not dogs! But some principles apply across species). The principle is, teach an incompatible behaviour. If you don't want your dog to jump on you when you come in the door, for instance, teach him to sit quickly for a treat or a pat or a toss of his favourite toy. If he's sitting, he's not jumping! Ditto kids and guessing -- teach them to read quickly and accurately -- reward a speed and quality of prosody that precludes guessing -- and the guessing will drop out on its own (or diminish significantly) because it does n't get any attention or reward. Visible tracking is valuable -- charts with points or stickers, for instance, as the student reads better, more and faster.
Very impulsive children may not completely overcome their tendency to guess, but you can make a huge difference in how frequently they use this strategy by changing what you pay attention to and reinforce. If you decide to count behaviours, like I did, you need an accomplice. The person instructing the child can't be the one doing the counting. An old rule of behaviour that we can't be reminded of too often is, you get the behaviour you pay attention to. So we need to set it up so we are constantly praising children for guessing-incompatible behaviour -- real reading!