Uhh, it depends who you ask, but the current science says that it's not the first thing on the list to take care of. Flat response is. Now, phase is not a bad goal by any means, it's just that you shouldn't focus so much on it that you incur other compromises that might be of greater severity.
Anyhoo, supposedly the Quad ESL is the only perfect speaker there is with phase. I've heard him talk about the square wave reproduction before.
http://forums.audioholics.com/forums/showthread.php?p=781319#post781319
He goes on to mention that he thinks that our research should be focused on a wide band driver to cover the greatest/all of the midrange (and then some, perhaps depending on xover slope), and that there are only two that are really good, one by BW and the other by PMC. That's what he says anyways, and none are available to the consumer as an OEM driver.
I've asked him about 1st order xovers, and he taught me about getting lobing tilt right with sloped baffles like with Thiel, but in the end, there are compromises of getting transient response, such as offaxis response. We are not so sensitive to phase, and one of his main points is that most recordings aren't in phase anyways, so what's really the point of focusing so much here if you're going to screw other things up. Offaxis response OTOH is very important according to the studies of Toole and others.
Here is his take on the difficulty of dealing with one thing in the face of a hundred things.
http://forums.audioholics.com/forums/showpost.php?p=781319&postcount=332