Consider both the HTTP headers sent for the page upon which the form is, and the HTTP headers for the page which the form POSTs to.
Major brownie points to whoever comes up with good data that'd let us fix the ever-occuring "LJ ate my entry!" problem when we show an error message on page 2 and erroneously expect people to be able to press back and change their data.
Of particular annoyance is that Mozilla's behavior changed recently, and I'm wondering if there's a spec out there (part of RFC 2616?) that I haven't read which Mozilla is now obeying.