Let us do some math, shall we? The assertment that dump succeed while skilled succeed is less than 20% is true if we look at only a single check situation over very many rolls. The actual chance is just over 11% (0.25*0.45). However the proposition that it might happen 1 in 5 times might still be very reasonable. If the dumper and the skilled is trying some task 5 times, the chances for the dumper succeeding at least once when the skilled fail is 45% (1-0.8875^5).
Moreover the assertion by
@Ruin Explorer was not neccessarily tied to a single check situation. If we for instance have a contested situation that only resolves if one fail and the other succeed (both succeed or both fail is rerolled) the chances of the dumper winning actually boosts to 21.4% ((0.25*0.45)/(0.25*0.45 + 0.75*0.55)).
You are correct in that higher DCs skew the math to the benefit of the skilled, but easier checks skew it the other way around. My take is that the proposition that stat is overrated in D&D 3+ is right, and that you should be carefull to try to counter claims about statistics before conferring with the devil first. (It is amazing what statistics can justify depending on how it is spun)
(Edit: corrected the last calculation)