For 6V AC.
You're right on the resistors and yet I tried that first and it didn't work in real-life. There's a lot of overshoot going on causing greater efficiency loss via shorts.
I created three circuits with a large smoothing capacitor and fixed wire resistor to compare the efficiency. Measurements by oscilloscope:
1. 4x 5819's - 5v - 5.5v (5.25v average)
2. semi active, IRF7103: 5V - 5.8V (5.4v average)
3. active, DMHC4035LSD, LM393: 4.7v - 5.4v (average 5.278v)
This was with much faffing to the above circuit, which especially requires two 473 capacitors between the resistor dividers and ground. This may be specific to the DMHC4035LSD characteristics. I found the LM393 was overly sensitive to Vin chatter which would cause false outputs to turn the FET's on. Looking at it, it would probably work to use a boost for stable voltage to the comparator.