Like you said in your email; NMOS doesn't work because you can't turn-off the device under no load and standard Vt PMOS will not work at such low Vdd. You need to use a low-Vt PMOS follower or a Natural Vt NMOS follower based structure. Maybe you can lower the Vt of the PMOS power driver or PMOS buffer to save headroom at high load or lower the Vt of the NMOS follower if you don't have the two components previously mentioned.
I've got two patents on how to play this game, of course you can't use them unless you work at TI or a company that has cross-licensing with them, but you may get some inspiration:
1. US Patent 6,861,832  
2. US Patent 6,646,495
Good luck.