First, thank you to @picokernel and to @contentjunkie for the work you've done on this.
One point of clarification, though. The more I think this through, the more confused I get. The miner throws its work away and starts over every 3 seconds, right? So how does it matter if the N cores are chasing 1 problem or N problems? Either way, I think they're only going to hit one block (at most) during the interval, aren't they? If N cores are chasing one problem, they've got N chances to solve it. If they're chasing N problems, they've got 1 chance to solve each of them. (N * 1) or (1 * N) - same net result. Not saying you're wrong, since I haven't read the code, but I'm just trying to understand.
Guess I'll have to break down and RTFM one of these days...
Last, FYI, I tried the windows compiled executable, but for some reason, I could only get it to use 1 core. It ignored my "mining-threads" setting in config.ini and the "-t" command line argument. Not sure if anyone else is having this problem or not.
Update: Clicked through to github. Now I get it. I hadn't understood that all cores are using the same nonce. So the miner really only has 1 chance to solve 1 problem with the unpatched version. Not N chances.