Intelligence is largely similar between all humans: we don't have actual boundaries. The normal intellectual boundary is artificial; even physical limits are artificial.
One of the most famous examples of the human artificial boundary phenomena is running. For the longest time, a four-minute mile was considered physiologically impossible. When the record was broken, it was swiftly broken again by another bloke a month later. Within a few years, everyone was running four-minute miles. It's now a standard, and the record is much lower than four minutes.
What held humans back from breaking the four-minute mile was not believing they could run a mile that fast. By not believing in the possibility, they trained themselves to assess those last precious seconds as the best they could do; they would push themselves, find difficulty, and assume this was as good as it gets. They wouldn't push themselves further because the exertion was interpreted as some sort of dangerous violation of what is possible, safe, or sane: it's hard because it can't be done, so the pain and exhaustion mean it's time to stop here.
In modern times, Olympic records are broken every year; mental mathematics are continuously improved; and humans at the World Memory Competition continuously break previous records for memorizing lots of shit in little time. Typists type faster, bicyclists bicycle faster, and IQ tests follow a sliding scale such that Einstein was kind of dumb and we've repeatedly revamped the Culture Fair and changed the baselines for the Wechsler. People are of the mind that anything that can be done can be done slightly better, and continue to progress by degrees over their predecessors.
What makes this progress possible--and what impedes it--is the form of practice taken. Time spent practicing has little to no impact on skill; it is the mode of practice that matters most. K. Anders Ericsson published a theory of Deliberate Practice: that a person practices in a goal-oriented manner with a focus on technique, folding in constant, continuous feedback to improve upon his deficiencies. In short: a person who simply tries repeatedly to memorize a deck of 52 cards will make small gains; while a person who reviews and notices himself confusing or slowing on specific cards will target those cards, correct the issue, and make rapid gains both small and large.
All of this brings us to a head about intelligence, and about the permanence of training.
Intelligence is a matter of creativity: a person must be able to apply knowledge to solve problems, rather than repeat back rote facts. Creativity is, in turn, a matter of knowledge: invention and inventory are the same; you invent by reassembling the inventory of your mind into new forms, dividing a problem into recognizable components and adjusting solutions to similar components so as to produce a solution. Knowledge is, of course, a matter of memory: you cannot know what you do not remember.
Memory is improved by technique. The primary considerations are meaningfulness: information is best memorized when it is organized (grouped) and attached to well-understood ideas. Images are immediately well-understood, and so visualization is used to convert complex thoughts into meaningful representations of known topics (i.e. a running duck--both "running" and "duck" are meaningful--can be visualized). Attaching sounds, smells, and actions makes a more vivid, accessible, memorable image; and complex techniques and systems such as linking, story forming, and mind palaces further aid in recall by providing indexing or association.
Synesthetes make concepts meaningful by attaching other concepts. Sound forms its own imagery, or numbers have their own smells. The mimicry of this is a core technique in memory improvement: speed card participants attach playing cards to images, emotions, smells, sounds, textures, and whatever else they can; while numeric memory is aided by a PAO system, converting numbers into people, actions, and objects. As long as the system is in use, numbers are associated to images, and numeric memory is drastically improved.
The 12-point IQ boost would be as permanent as the effect. If the effect remains in some form--if the sensations are lost, but the internal memory of sensation is associated with the things being examined--memory will be improved. If memory is improved, working memory can more efficiently process things, reducing confusion and improving the accessibility of complex tasks. All that is required is a permanent association between the things being synesthized: a person looking at the number 27 might no longer taste lemon, but he may be immediately reminded of the taste of lemon as an idea, which will more strongly stick in the mind. When you ask him numbers, he will remember tastes and smells--themselves not having been experienced, but having been remembered when he first examined the numbers, which is just as good--and will remember the associations that bring those together.
So it's quite possible.