Yeah, I think ArdWar has the right solution there, and the circuitlab.com seems to agree. Even though their symbol doesn't have a body diode, it seems to behave like it has one, and as soon as the output voltage at the Drain start being 0.7V higher than the Source, it starts dropping the voltage. I tested it by changing the 3.3v source to, and as long as it was higher than 4.3, everything was fine, but it started conducting below that.
The only difference is, I would need 3 MOSFETS back-to-back instead of 2 for the 1.7V (5V-3.3V) Drain-to-Source voltage differential, and ArdWar's design had it on the right (5V) side, whereas it needs to be on the left (3.3V) side.
I'm attaching what seems to work, at least on the simulation.
Wow, thanks everyone!