- this is a serious question that bothers me; I can't see how this kind of selection gives any kind of representative sample.
Yes, Alexa information is pretty bad. Yes, it is easily faked. Yes, it isn't representative. Yes, browser toolbar usage is probably declining.
However, that doesn't mean that Alexa information cannot be used as an indication of big traffic trends considering there is a difference between absolute traffic levels and relative changes in traffic levels.
To give a different example. Let's say that you have a website with decent traffic levels and you collect browser usage statistics (e.g. 50% IE, 40% FF, 10% Chrome). Of course those are not representative for global browser usage. But if Chrome usage doubles for that group, it can be used as an indication for a trend in global browser usage.