Thanks--this is fascinating. Either of those works much better for me than the original. Moving the top of 3 down helps also--I should've noted that something of the same happens in the panel-edge example I posted, 2.6.39 (which perhaps makes it not as good an example on the sheer power of panel edges).