ISO is a measurement of your camera’s sensitivity to light; lower numbers are less sensitive, and higher numbers more sensitive (ISO 200 is twice as sensitive to light as ISO 100, for example). However, the more sensitive a camera’s “film stock” — or in this case, the digital equivalent — the more noise in the image. With film stock the relationship between sensitivity and noise is generally linear (more on film speed and its digital equivalents), but Canon’s implementation is somewhat unique.
While the cameras have dozens more ISO settings, it turns out that settings in between these values are artificially boosting (or lowering) the camera’s exposure compensation; you don’t actually gain any highlight headroom (e.g., ISO 200 is just an exposure-compensated ISO 160). While the difference between “native” and exposure-compensated ISO levels is not drastic by any means, if you’re shooting in a difficult situation without any lighting options it can help to know that a shot at ISO 1250, boosted in post to the same perceived brightness as an ISO 1600 shot, will be cleaner than the native ISO 1600 image. You can see this on a noise levels graph and it also apparent in these 5D highlight tone priority tests.
What is true is that the noise levels at these particular ISO levels are lower. However, as has been pointed out by a couple of helpful commenters, this is actually because the camera is electronically dampening the exposure, giving the image 1/3-less stop of noise whilst also giving 1/3-less stop of dynamic range. Basically: it’s a wash. I’m leaving this section here in case someone read it before and was going out of their way to select 160, 320, 640, etc. — or in case someone read this “native ISO” information elsewhere (as I did originally). Move along, there’s nothing to see here (except a bit less noise and a bit less highlight retention).