I'd say that "concept first" is often a good way to approach RPG development--provided that you have a
sufficiently rigorous and narrow view of what encompasses "concept". The more lazy and broad you get with "concept," the less that will work.
Really, once you've got a good concept nailed down, then you can do fluff first, mechanics first, do them in tandem, alternate, or any number of combinations and mixes. Then on the next concept, you can go do it some other way. The problem is that most people write RPG concepts like middle managers write "vision statements".
For example, "Illusionist" is not a good concept--even with the implied baggage/assumptions from prior versions that the word may invoke. "Illusionist" would be one good key word in any number of single-sentence concepts. Think about those prior version assumptions, and you can probably come up with such a good sentence.