If the test is using the standard IQ measurement scale, then a 20 point variance is gigantic. Were they the same test type though (eg were they all stanford binet)? I also suspect that the tests you did were experimental (and failed) or not applied correctly as that variance should not happen.
I also agree that the results people are getting from that online test are really funny, especially when you factor most of the world is around the 100 mark (100 is dead center average) and only some where around 0.1% would be scoring near 140 (I forget the exact % but here is the curve if you want to see
http://en.wikipedia.org/wiki/Image:IQ_curve.svg) that that online test is spitting out so many super high iq values completely proves my original point.
Personally I'm not a fan of the stanford binet test, though it may make for a decent representation of intelligence in the western educated world, it fails miserably with anyone not educated very well in the west.
Personality tests are another subject all together. I do think Eysenk's 2F test can be somewhat useful (though limiting as its only 2 factors, I don't think personality can be boiled down to a sum of 2 factors), the 5F OCEAN test seems reasonable as well in principle. I would however stand by the MMPI v2 (Minnesota Multiphasic Personality Inventory) test. That one has been proven time and time again as an effective diagnostic tool (for the western world anyhow). But as for the first 2 I mentioned, it is based to a great deal on theory.