I know, it's not a "button" if you don't feel that you can _press_ it. Check what I mean in this image (mod using box-shadow): http://i.imgur.com/ZyiH7AJ.jpg , a few visual clues can help a lot to the user.
But of course there are alternatives for visual clues that still look flat-ish: http://i.imgur.com/jXN2tpe.jpg
You should not generalize from this example that flat buttonlike constructs are as usable as normal buttons in other contexts.
Also while I personally don't like the flat design, I believe it can work with a caveat: you will need to be much more careful in the design as you offer less conventional cues. In this I believe a design that offers more affordances is less prone to small mistakes.
Couple that with the lack of a form.
I'm not saying that flat buttons are superior, or that 3d looking buttons are worse. Simply pointing out that context matters, and even your opinion is worthless. That it looks like a warning message to you is indicative of other problems, not merely the lack of 3d.
The new Blackberry OS is like this with gestures. Sure, you can spend 20 minutes moving your finger(s) in every way imaginable to learn how to use it because there's no chrome (or hardware buttons), but it certainly doesn't instill confidence on the first use. Instead, it presents itself as a challenge: "learn how to use me, I dare you."
Not only that, it requires a mouse user also to hover over all possible elements to just know what's clickable. Kind of like some of the first graphical adventure games.
Or imagine a website where a link only turns blue and underlined as you mouseover it.
"Input" versus "Indicator" is a big problem in FlatUI that needs to be solved, there is no shared visual language yet to signify which is which, and it needs to be possible by just looking (and preferably other modalities as well)