Away from my iMac until tonight, but the next thing I'm interested in running is the same test(s) on a 1080p external display.
I'm thinking that once you get into the business of updating 14M physical pixels every 30 seconds or so some of the relative advantages of these GPUs are diminished.
Well, just like full-screen vs windowed, which was a clue I guess, the Unigine Valley Benchmark is basically insensitive to how many physical pixels are getting updated, at least for the M290X.
Here are the results for an external 1080p display: