Not sure what your question is. A hash is a hash is a hash. For pools? There is no advantage to having the same card generate multiple work threads, except that some people suggest that you might squeeze a couple extra Hashes out of the parallelized GPGPU structure. I don't really see it myself.
Yes, this was essentially what I was wondering. Over the long haul it wouldn't really matter I supposed, but on the 0-minute "lucky" block solves where a multi-GH/s worker might only contribute a dozen or so shares before the block is solved, it could make a significant difference if indeed splitting the hashing power up into 2 or more workers makes a diff.