Ah, you see? You have benefitted from Avbug's wisdom, already.
I vote for cool and light, too. It gives me better options, including whatever single engine climb I may need for obstacle avoidance.
Since published Vmc is something of a "worst case scenario", the actual speed where control is lost will likely be lower. I would only experiment with the region below redline with a brick s***house directly in my path, and immediately pitch back to blueline, if possible. Just as in the Vmc demo, at the moment you sense a loss of directional control, you MUST pull the power to avoid the rollover. You may survive if you land cockpit-up. You won't if you land cockpit-down.
Summary: Land if possible, continue takeoff if not possible, and know what performance may be conservatively expected before you taxi onto the runway. You might have to wait until later in the day to safely depart in a light twin.