I disagree 100%!
By doing that you are just asking for mixdown summing problems in the end, defeating all that hard work. You aren't leaving yourself any headroom at all... which means you'll have to turn everything down--defeating the purpose AND increasing noise floor.
My rule is no more than a -6db peak, RMS around -12 to 15dbfs. Keep in mind that for every doubling of track count you should reduce your inputs by 3db to get a clean mixdown.
If I'm doing 32 tracks I aim for a -9db peak, RMS around -15dbfs... if I *were* to go to 64 tracks I'd be looking at -12dbfs peak, RMS around -18dbfs and so forth.
Recording as 'hot as possible' just doesn't make sense. You gotta capture those fast transients that may not measure on your meter/peak lights. Otherwise you're going to be playing digital doctor on the waveforms... which is double plus unfun.