I can create the problem on the ground by running the engine at WOT for longer than two minutes. The data logs show the EGTs are nearly identical until the rough-running starts, and then they are about 30 degF apart; the fuel flow is erratic during the rough period, and varies from 5 to 7 gph (4.9 gph is the flow when it's running properly); and an ignition check at 4400 rpm (and running roughly) has the same drop (about 150 rpm) for Left and Right. The electric fuel pump was not running during this ground test.
I've weighed the floats, and all four weigh 3.1 grams.
My mechanic is perplexed, but still thinks it's fuel problem, and likely a carburetor problem, but he'll have to visit the plane next week to pursue a solution.