This might be very easy to fix. I'm not sure how easy it is in our CSS system to have CSS properties which only explicitly inherit.
I have what should be a fix in my local tree, but something's not quite right. Still investigating.
Actually, I believe the test is bogus. http://www.w3.org/Graphics/SVG/Test/20061213/htmlEmbedHarness/full-pservers-grad-18-b.html the test reads: <g stop-color='red'> <linearGradient> <stop stop-color='inherit'> However, stop-color does not inherit. So <linearGradient> ends up with the default stop color (black) which is what the <stop> element then inherits from. the CSS inherit keyword does not cause the style resolve to jump to the grandparent, especially since linearGradient is a styled element... This is just a bogus test. Oliver or mjs (or whoever has contact with the SVG working group) should complain to them.
ed_work in #svg agrees with our interpretation. We expect the test case will be changed to correct this error in the next release.
As Eric said, probably wrong test, closing for now.