It's easy to show in detail how a big, friendly and accessible label makes a field more usable, but in the context of a large, complex form, the overview of the form structure and completeness of the data already entered are also important. Too prominent labels and too much help text can distract significantly, also for those who are visually impaired or are relying on screen readers.
My conclusion is that some form of two-phase, view/edit approach seems to work best for large forms. Labels and help are minimal by default, but the parts or "blocks" the user is actively editing gets more help and meta information added while it has the focus.