Just to confirm:
As shown in the image below, on track 1, I increased the gain to the maximum to force the clipping of the audio.
And on the Master Bus, I turned the volume down, so on this bus there was no clipping.
The question is: was the audio effectively lost on track 1 (before it goes to the bus), as indicated in red?
Or not, nothing is lost because I lowered the volume on the bus to compensate the excessive volume on track 1?
In other words, should I worry about audio clipping on the original tracks or just on final bus?
 

The audio is definitely clipped when only one of the plugins has reached more as 0dB.

I wouldn't worry about using the gain too much unless you are feeding (for example: a VST plug in) that has no input.
In short: The Gain knob is the input section of the desk, the fader is the post section.
Always make sure your Master bus is set to 0db/Unity. Turn down the faders to make up for the overloading.
 
Theoretically, each track adds 3db in the over all mix.
If it were as easy as sending 1 track to the Master bus, this would be cake/easy.
Because there would be no difference. but as you add more and more tracks, each track will add 3db to the over all end point/Master bus.
 
In conclusion, turn down your faders and check out your signal path so when you gain stage your not overloading anything.

Once you cause/record distortion you sill hear it. If the destination buss is still in the green, then you will still be carrying the distorted sound but it's not turned up enough to cause new distortion on the destination buss.
 
That's exactly what your screen shots show. The source is overloading but the buss isn't. But the output from the buss will have the original overloaded/distorted sound.

Your track is "clipping" so it would be best to turn down this track (gain first to 0, then the track fader) to get it out of the red

Within SONAR you may or may not actually "hear" the clipping but it is good practice to avoid clipping on the track, even "in the box"

The most likely place you will actually hear the effect of clipping is when you drive a plugin or VST too hard, especially in the pro channel where you go from plugin to plugin - keep an eye on the clip indicators and use gain staging to ensure the volume coming out of one plugin does not overload the next one (which is easily done)

You can insert an empty "dummy" FX Chain to give you some level control at any point within the Pro Channel to help keep things under control

Sometimes it helps to think of digital audio as simply data. Changes in the value of each sample can be recorded up to 0 dBfs. Digital overs ("clipping") happens when the numerical value of the sample exceeds that threshold, at which point all larger values are lost or more accurately rendered as the highest value that can be represented. The fluctuations in the value of the sample are what carries the audio information, and over values lose that information. There is no way to recover information lost at any stage of the process, so clipping at any point is going to be carried forward in all subsequent operations. 

I did a test that answers my own question (and probably differentiates what has been said so far).
If Sonar indicates an audio clipping (red) on the source track, this DOES NOT MEAN THAT IT WAS DISTORTED ON THE SOURCE TRACK.
As the images below can prove, even Sonar indicating audio clipping on track 1 (which I purposely increased + 6db on the volume), it is not actually clipped internally, and can be completely recovered if I compensate the volume on the bus (which I deliberately decreased -6db in volume, to compensate).
 
1) I created a WAV file with a single sine wave, normalized to -3db:

 
2) In Sonar, I increased +6db, indicating audio clipping on track 1 and also on Master Bus:

 
3) The bounced file naturally shows the clipped audio (I manually normalized it to -3db for visual comparison purposes only):

 
4) Now, In Sonar, I reduced -6db on the Master Bus to compensate (and avoid audio clipping on the Bus):

 
5) And the image below shows that the original audio, even though shown as clipped on track 1, was fully restored back to the Bus, only compensating the volume down.

 
This way I can understand that the red indication of clipped audio on the original track does not actually clip the audio in the source, preserving it to the end (which is a good thing).

I've understood the (32 bit) floating point bit rate somehow eliminates the down sides of going to red as long as you're inside SONAR. When and how does it come to the picture?

Interesting result.
 
I presume one of the Bakers can confirm your findings definitively, but I assume this is due to the fact that the audio is routed internally at a higher bit depth than the actual bit depth of the project so there is additional headroom until the very last output stage.
 
AMB

 
Here's the short answer. It's not 100% technically complete, but that's why it's a short answer.
 
Digital audio does not clip within the DAW, up to and including the master bus fader.
 
When a meter indicates clipping, it's predicting the future; no clipping has actually occurred yet but probably will happen once the data leaves the DAW.
 
So your intuition is correct: you can indeed turn the gain down at the master bus and erase any previous level sins. 
 
However, there is one big caveat. The data passes through many individual processors before it gets to the master bus, and you can't know with certainty that none of them will be thrown off by positive dB values. In fact, experience has taught me that many plugins don't handle it well. The result can be a subtle degradation that you might not even notice until you bring everything down and suddenly your mix sounds better. Sometimes, the degradation isn't subtle at all.
 
Bottom line: you're better off keeping levels under control at every stage.