Didn't intend to come off as overly negative, and your data may be better than I gave them credit for (although most web statistics are only usable by marketeers and others less interested in reality than they are in pretty pie-charts; self selection effects tend to run rampant). It's mostly the assumption part. For instance, my normal viewport tend to be large enough for the page to look as intended, yet 'normal web usage' (i.e. opening my bookmarks) will break it. Navigational elements shift position, margins squeeze together and impair readability, I may have to scroll (horizontally!) to keep navigating the site, and I have to spend a couple of seconds 'relearning' the page before I continue reading...
All minor (and certainly fixable) flaws, but exactly the kind that impair web usability. A key to successful web-navigation is pages that are instinctively 'experienced' rather than read when it comes to the common elements, and right now the design fail at that. One other thing to look at in that regard is, for instance, the location and distinctiveness of the navigational tool for multipage content. Any departure from 'the conventional way of doing things' should be undertaken with care.
IM(NS)HO, you're wrong on that one. Take your 90%, for example: Alienating 10% of your users 100% of the time by design is a really, really poor move in and of itself, but that number may just as easily be interpreted as annoying 100% of your users 10% of the time, which may be even worse.
Oh, I'm certain.
[Insert mandatory no one ever reads the frontpage content anyway joke here]
