The Problem with Type Statistics
I've always found it interesting whenever someone mentions a statistic relating to an MBTI type, such as "ESTPs are only 5% of the population" or "INFJs are the rarest type." I also find it interesting whenever people try to refute a type statistic - for example, I may mention that Niednagel believes that ENTPs are the most common type in America, and people often respond as if there's no way that's true - even though they can't provide sufficient evidence to prove or disprove his statement.
Even within an individual, it can often be difficult to type oneself accurately. This becomes an even greater problem when trying to accurately type a large population. Here are some of the issues involved in attempting to accurately type a population:
Sampling Error
Of course it is impossible to type every single person (whether through having them take a test or having someone evaluate them). So, just like any other attempt to collect demographic information, type statistics would be derived from a sample, or a portion of the population. Of course, it is very difficult to get a sample that is large enough and diverse enough to represent a large population.
For example, people who are interested in personality typing tests often identify themselves as iNtuitives because they are interested in theories such as personality theory; obviously, this can skew results.
Methodology
The two main methods of determining someone's type are having someone take a test and observing someone's behavior. Both of these methods have their flaws. Biases, as I explain below, can influence both of them. Understanding the test questions can be an issue, as can having an accurate sense of self. As far as typing others goes, one can observe behaviors, but not necessarily understand another's thought processes or their past history; there are a lot of influences on behavior besides personality type.
Biases
Biases can influence both test results and behavioral observations. Books have been written about sources of bias, so I won't go into detail on every possible source of bias. And in fact, someone else has already written about biases in the past, so I'll just direct you to that post:
http://personalitycafe.com/myers-briggs-forum/2757-caution-know-your-bias.html
(Credit to NephilimAzrael for the above article.)
The Most Important Question of All . . .
When it comes to type statistics, the most important question to ask is, what are you going to do with this information? Why do you need to know how common or rare a certain type is? How is this going to change your personal development? That is something you have to figure out for yourself.