Top of my head thoughts...
IMO, the Google Adwords External Tool has the largest sample size, from actual Google Searches, and gives the most granular data. But you have to learn how to use it. Eg, you do not want to have everything on Broad Match if you're using it for organic suggestions (as opposed to PPC suggestions). Used properly, it can give you excellent suggestions with surprisingly good information about infrequently searched terms. Geo bias is adjustable in the tool. It has an internal algorithm skewed by your input data that affects what data it returns, and that can be helpful or confusing, depending on what you're trying to find out.
The free Trellian Keyword Discovery Tool gives data accumulated over time. I believe it uses Toolbar data, so it tends to ignore certain kinds of searches, like B2B searches in market areas that apply only to large companies (ie, it's not likely that searchers in those environments will have toolbars installed). I've found it very good for predicting relative number of searches in B2C areas. The numbers it gives don't give actual monthly searches. I see apparent geo-basis on some placename based searches. I've heard they're working these out. I like the tool, which has actually helped me spot anomalies in the Google Tool.
Hitwise is reputed to have a very large sample size, I believe also uses toolbar data (maybe IP data too, I'm not sure), and for the same reason is likely to be skewed in the B2B area, but is said to give excellent information for those in the B2C area. It's a very expensive service, prohibitively expensive for many.
WordTracker has its following, has a beautiful interface, but has more whistles and bells than adequate data. It derives its search data from a very small sample size, pulled from actual searches on a small meta search engine(s). It's useless for identifying infrequent searches... can be very good for frequent searches. Because of its small sample, it multiplies its errors up when it scales its results to take into account market size. Its demographics are definitely not the same as Google's, and I'd be careful about using it as a predictor of certain kinds of niche searches.