My preference for moving AMs to an expanded FM band (replacing TV 5 & 6, for example - 76 to 88 MHz) would be to mostly move the smaller AMs that have relatively small coverage footprints.
For example, if...
a particular AM (low power, high dial position, inefficient antenna, poor ground conductivity and/or crowded channel)...
cannot be detected by a DXer with a high-end ($1M+) 530-1700kHz-only (so as not to "waste" money on other bands) communications-grade receiver and beverage antenna, day or night (even factoring in once-in-a-hundred-years DX openings favoring the station - assuming EVERYONE else is off the air, there's a massive power grid total failure blanketing a 12,500-mile radius, your receiver is on battery power and the station's on generator), ...
but an FM at the same distance with the same power could be as clearly heard as a properly-working FM HD on a cheap pocket Coby radio (one with a built-in speaker) WITHOUT an antenna (read: headphones plugged into the headphone jack), ...
then I'd say move the AM to FM - among other things, the coverage should be improved, and it would clear up the AM frequency so it's not quite as crowded.
Before you think of moving ALL the low power AMs, though, I should mention that someone in Oklahoma recently heard a southern California station that's supposed to be only sending 12 watts his direction, and I've heard of TIS and Part 15 stations (when the frequency is clear) being heard thousands of miles away. (Well, maybe not thousands of miles for the part 15s, but hundreds.) That wouldn't, in my opinion, mean the station is "impossible" to hear. Sure, it might be rare, but there's still a chance. I've also heard the TIS's at DFW from southern California when the expanded AM band was first starting to be used.
Or should I not be trying to figure out how to accomodate the DXers that are looking to log sub-watt stations from thousands of miles away?
Also, S-Cat, you mentioned that in a "quiet" area a 50 µV/m signal could be heard fairly well. About what level signal would you say, under the best conditions (preferably groundwave to eliminate the variability of skywave), would be right at the noise floor, ranging from carrier barely detectable to program material barely discernible with good quality headphones? (I'm assuming a "fairly well" heard signal would be something like "arm chair copy" (able to listen at a moderate volume on a portable with a built-in speaker while washing dishes by hand across the kitchen, with the water in the sink masking the noise in the radio), or at least a good enough SNR as to be the type of signal that would trip the seek/scan on a typical consumer radio if it was receiving it that strong. A "clear" signal to me, at worst, has the RF noise floor far enough below the receiver's inherent noise floor so as to be undetectable.)