yeah this is an interesting topic and one that gets complicated further by all of the different db reference scales out there as well!
Technically speaking, when recording in the analog domain (i.e. on tape back in the day) you always got distortion and noise added to the signal. If you recorded too low, your signal was distorted/ coloured by tape hiss and if you recorded too hot, tape saturated distorted/ coloured things. 0VU was setup as the average signal level that provided the best signal to noise ratio with the least noticeable colouration. Transients would go above this level and would get saturated by the tape, but this would be less noticeable than saturating average levels.... unless that was the desired effect.
Digital recording is slightly different though. The noise floor of digital equipment does vary from system to system but it is far superior in that respect to tape, so is typically regarded as a non issue. Technically speaking, digital also records clean right up to 0dbFS. As soon as it goes over, however, it sounds horrible! There was the adage of 16bit recording which prompted people to record over -6dbFS to use all of the available bits but this leaves little headroom for any signal variation unless you are using external compressors and limiters to tightly control things. 24bit recording has since, squashed this mentality.... or at least, should have done!
Even if recording mediums have changed over the years, recording requirements haven't. You want a healthy signal above the recording medium's noise floor, with enough headroom to allow for performance variation and peaks being well away from any 'ceilings'. You also want to work within the operating specifications of the equipment in your signal chain. This last bit has been the issue I think as when people were working with purely analog equipment, it was straightforward. In one sense, when working with only digital it was/ is, straightforward. It's when people interface analog gear with digital equipment that the confusions crept in. They both work with difference scales and reference levels and finding common ground is the grey bit. This is complicated further by the different requirements of recording and in turn, mastering.
To help simplify things a digital equivalent for 0VU was found. In europe, 0VU = -18dbFS and in the states, 0VU = -20dbFS. VU meters, due to their design, responded to average levels, therefore, the RMS of your recordings should be looking at -18dbFS (in europe) allowing 18db of dynamic range... when recording. Peaks probably shouldn't go too far over -10dbFS.... but this will also depend on how dynamic the performance is. If you think of vocals, some songs have whispering bits alongside shouty bits and as a result, would be very dynamic to record.... but you would probably be wanting to ride the tracking volume in that scenario remembering the essence of recording in optimising signal to noise and headroom.
This is what most recording equipment is designed to accommodate hence the comments about convertors running in their sweet spot in this range. If you are mixing multiple signals in to one, like in most typical music arrangements, each channel will be turned down somewhat in order to make the overall output cohesive, balanced and at a sensible level.
Recording at lower levels and turning up the volume of your speakers is probably the BEST advice anyone has ever given me.... knowing why helps clarify it all too :)