I thought of two possible explanations (besides "It sounded good at the time.").
First is the strong tendancy of mathematic progression that early 3E products seemed to have (a tendancy I really like, FWIW). In 3.0, IIRC, rogues were the best. Semi-skilled classes (ranger, barbarian, monk, bard) were half as good as the rogue. Unskilled characters were half as good as those. In 3.5, they just changed it from geometric to linear progression and added a forth classification.
Second, there's an optimal spread to the number of different "skill tiers", what the difference between each tier should be, and to how many actual points each tier should have. In 3.0, they decided that three tiers were appropriate and in 3.5, it was 4 tiers. They also decided that one skill point wasn't enough (thus 2, min) and they wanted to stay under 10 (so 8 or 9 max). Going with a 1 point spread doesn't provide meaningful distinction between each tier, and 3 points is too much.
They could have just as easily gone with 3, 5, 7, and 9, but then you'd be asking why the skill points only come in odds. In the end, evens were probably chosen because counting by twos (or basic doubling -- pick your edition) is a pretty basic math skill.