http://www.w3.org/TR/css3-transitions/#transition-duration-property specify that a negative value for transition-duration should be treated as 0s. It's not the case in WebKit today. I wonder if the spec is accurate, I remember it was discussed not so long ago in the CSS WG. My quick test in js (not actually looking the rendering) shows that Opera returns negative values when querying the transition-duration on the computed style and Firefox seems to just drop the negative sign, so -10s would be a 10s.
I really should read the editor's draft http://dev.w3.org/csswg/css3-transitions/#transition-duration-property has been updated to "A negative value for transition-duration renders the declaration invalid." so WebKit behaves correctly.