What hasn't really been stated fully here is WHY running cooler in 24bit is OK versus 16bit. Remember back in computer school about how additional bits gives more information? Well, I don't know the exact numbers, but in 24bit, each sample has an order of magnitude more information recorded than 16bit.
Running hot in 16bit was a little bit more important for rendering the sound information on the media accurately as it was recorded. Since there's less information, there's less lattitude with what you can do with that information in post. In that case, the general concensus was that it was worth the risk of clipping to run hot because alot of people felt that the sound tended to be better on the final product when they didn't have to bump the levels 10db or more in post.
In 24bit, there is an order of magnitude more information. Most everyone agrees that the sound information captured with levels running conservatively and then bumped in post sounds no different than if you run hot to begin with. So, the logic therefore is to run conservatively for the sake of avoiding any potential risk of clipping.
Regarding, the concept of amplifying noise, that is a concern for sure, but as someone has explained, the noise floor for most of the music that we record is such that it will not be heard, especially with modded low-noise components in the sound chain. Bumping the noise floor 10db won't be heard anyway. I suppose if all you record is chamber music, this would something that you might want to check out on your rig, but I have the feeling that the use of high quality, low noise components will keep the noise floor sufficiently low that using the '10db bump in post' strategy would still work out OK.