Except that when it comes to usability, using skeuomorphism tends to confuse user (note that it happens in some cases), because it implies that something will behave in a certain way like a real world counterpart, while that piece of interface might not.
I liken this stance of skeuomorphism to Douglas Crockford's stance on JSLint. To parahprase him "if there are two ways to do something and one of them can be a terrible source of conusion, use the other". This was originally meant for programming but it can apply as well to designing interfaces. So using skeuomorphism isn't immediately bad but it's a red flag, like saying if(a=b) in C.
I liken this stance of skeuomorphism to Douglas Crockford's stance on JSLint. To parahprase him "if there are two ways to do something and one of them can be a terrible source of conusion, use the other". This was originally meant for programming but it can apply as well to designing interfaces. So using skeuomorphism isn't immediately bad but it's a red flag, like saying if(a=b) in C.