I realize this is quite an old question, but I can provide some more detailed information. My source of this information is a combination of Kindle log files and de-compiled Java classes for the ReadingTimer package.
There is a lot of detail behind the statistical methods they use to avoid incorporating outliers into their calculations that I have uncovered - but I won't go into that detail here.
Basically, every time that you turn a "page" on a kindle, they capture the time you spent on the page (in milliseconds), and the number of words on each page. They call this an "Interval".
They calculate an Interval words per minute (WPM), and if that calculated WPM is > 900, they exclude that observation from the sampling methodology they use. If the WPM value is < 900 - well then it gets complicated. It's nowhere near as simple as just averaging together all of the IntervalWPM observations. There are outlier discovery, normal distribution, z-score testing and other statistical methods utilized.
But, from a drastically oversimplified perspective, those Interval WPM's are used to construct a statistically robust TotalWPM. That TotalWPM is, roughly, the average (but not really the average) WPM for the specific book you are reading.
So, bottom line, they do actually have a limit on how fast someone could read: 900 WPM - which is about 3.5 times more than the commonly recognized average reading rate of a random individual (native English speaker reading English).
And on the flipside, they do not technically have a limit on how slow someone could read. What they have is a limit on how much slower someone reads any specific page relative to their normal reading rate (which is tracked across books by using something called the GlobalWPM). So - if you start with a fresh kindle and only read 2 words per minute, that will be your acceptable baseline rate.
I have, of course, drastically oversimplified what's going on behind the scenes. The ReadingTimer code is fairly complex, and with how much information is lost between a decompiled Java class and the original Java code - figuring out what its doing is quite time consuming.
Hope that helps.