I'm mixed, and have always been mixed. My first reaction was primarily negative: "Is it really time yet? Even if they pull something pretty nifty, they could have waited another couple years." This changed as they started talking about the mechanical weaknesses of 3E: things that I had believed for years. Heh, it looks like they might have finally hired a math major. Good for them!
The new flavor has been disappointing in the extreme: the infamous Golden Wyvern (and other Wizard traditions) was, frankly, absolutely ridiculous. The core books are designed to be setting neutral; they said so themselves. Come on, I was so happy to see the stupid Druid Organization rules dropped from 2E, so why on earth did they have to add more setting-based things in core books! 3E finally got it right, and then they go and screw it up again. Figures. The use of Dragonborn (and the general shift away from Tolkienien Fantasy) was another issue. There is some flavor inherent in the races and classes presented: a world with only humans and no magic is dramatically different from Harry Potter because the rules about what can and cannot happen or exist are fundamentally different. That flavor should then try to please as many people as possible. Well, Dragonborn aren't part of that, but things such as elves and dwarves exist in most fantasy worlds, and serve as a sort of default. Putting them in the PH dictates there existance to DM's without the time to rewrite fundamental options, but most DM's would put them in anyway. Dragonborn, not so much. I would love those in a supplemental book, but I wanted them to stick to the basics in the core rules.
There have been some very pleasant surprises, too. The matching defense AC and attack modifiers based on level are so much more mathematically sound than the previous system, which always seemed to work (or not work) because of playtesting and some luck, but not because anyone actually thought about what they were doing beforehand. I thought that it was too much to hope for them to do the same with attack and AC, and use a one-time constant bonus, and lo and behold they did it! Hallelujah! Removing the crit confirmation roll (if it is indeed gone) is bad for exactly the same reason: crits (without that rule) tend to be difficult to balance because no one knows how often they occur.
I have some concerns with multiclassing that they will, I hope, address. The one-time static bonuses (to defenses/attacks and also trained skills) seem like they make the first class matter most in the case of multiclassing. I hope that they do not make multiclassing feel like the first class is always the primary one. Grrr. Maybe the "<class> training" feats fix this. I cannot think of a mathematically elegant solution (other than those feats) that is reasonable to compute and not abusable (or, on the other hand, that doesn't punish people for multiclassing).
I am now optimistic, but a bit worried. They've done some wonderful things that I wanted to see done for so long, but never dreamed that they would actually do. They've also rewritten some flavor that was way too specific and sometimes plain bad (or plane bad... get it? The Great Wheel). This would have been all fine and dandy, but they added more specificity for no apparent (good) purpose. The way it looks right now, I'm probably getting 3E. This has changed over time, but not in any steady pattern. We'll see how it continues to change.