Lots of people seem to be weighing in on GRE's lately, including two of my favorite bloggers, FSP and Janus Professor. I'll add my perspective to those of the bloggers and commenters out there.
The qualifications that make my opinion worthwhile: I took the general and physics subject GRE's in 2001; I'm a PhD candidate with a reasonable amount of information about what it takes to complete a PhD; I served on a committee to admit PhD students to our university a few years ago.
Links where I will pull some information about scores and percentiles:
The quantitative section is all about doing 9th grade level math relatively quickly. Many (most?) of my physics PhD classmates received a perfect score of 800 (94th percentile). I did not, because I ran out of time before the last ~2 problems. My <800 score really does indicate that I'm a little slower at basic math than most of my physics peers. Not that I don't know it just as well. In many cases, I find I know it better. But I find it really, really hard to force myself to do it faster. I'm addicted to the double-checking.
Anyway, in most circles (including on the graduate admission committee at my university) an almost but not quite perfect score doesn't alert any red flags, but I have occasionally heard people express that no one in a quantitative field should be allowed to continue without that perfect score. Those people are wrong. I don't really understand why the quantitative section is so focused on doing such low level math so quickly. For testing quantitative reasoning, it seems more useful to throw some more complex problems into the mix (not necessarily even requiring higher level math, just higher level reasoning). If anyone knows of a justification for sticking with the quick, low level math problems, I'd love to hear it.
The verbal section tests verbal skills via analogies, antonyms, synonyms, and reading comprehension. The hardest thing about this section (from my experience and according to everyone I've talked with) is the vocabulary. The vocab is really high level, esoteric, and a lot of words I had never encountered. "Esoteric" is a GRE word, by the way. I studied vocab for this section for an evening, and it helped me a great deal on the exam. As a graduate student, I've since encountered a few of the vocab words from the GRE. I suppose that it is helpful to already know the meaning of these words when they are encountered, and saves a little time to not have to look them up. I don't think it's wise to use most of these words in writing, because the vast majority of people will not know what the word means, so the words will detract from the clarity of your writing. But I guess one could make that argument for all words above the high school level? Anyway, personally, I feel that using such high level vocabulary takes away from the point of the test. The point of the test is to give graduate admissions committees an idea about your verbal reasoning skills. Yes, vocabulary is a part of that skill, but vocabulary is the part that can be most enormously improved by a short-term cramming session. Do they actually want to test our verbal reasoning skills? Or some combination of verbal reasoning, plus a test of whether we were willing to cram vocab in our brains for a short time beforehand?
As for our physics graduate admissions committee, verbal weighed in at about 1/2 the weight of the quantitative score.
I know the relevant section these days is the analytical writing section. From talking with people and from my days on the graduate selection committee, this section requires writing a couple essays. The analytical writing scores are...bizarre. I saw many applications with very good (even excellent) writing in the personal statements and extraordinarily dismal (<4) analytical writing scores. People who I know to be good writers (because I've read several things they've written) have told me they received low scores.
On the graduate admission committee, we ignored the analytical writing scores completely.
As an aside, I have to say I looooooooved the old analytical section, which had logic problems. I would do that shit for fun. And I scored in the highest possible percentile on it, just to toot the old horn a little. Did it indicate anything? Again, I think most of my fellow physics PhD peers did really well on it, but I feel too biased to really make a strong statement, and it doesn't matter anyway since it no longer exists.
I'll have to make a separate post about the physics subject GRE...