But people use those undocumented quirks in browser
implementation. I don't think we could ever banish
reliance on such non-standard features, so maybe the best
way to deal with them would be to incorporate them to the
RFC in substitution of the original spec (I'm not talking
specifically about this one quirk, which you say is unsafe;
BTW, would you dwell into why it is so?).
PerlMonks itself, doesn't it rely on Netscape's redirect
after POST behavior?
Also, consider that Perl is a good example of the
implementation being also the standard, a system
that has worked pretty well, IMHO.