And finally imagine you want to record performance, that involves both heavy synth parameters changes, and huge random changes. Glitch synth, for example. You can be very satisfied by what you have record, but you can never get the same result again if you only record parameters. And the reason is that random changes can influence your performance, and with other random seed you performance will have no sense.

Bristol_Jonesey
1: Record. (how long? 3 minutes? 10 minutes?)
2: Freeze. A freeze/unfreeze takes seconds

Freeze needs Listen (in case of random sounds), and Record can be combined with Listening, so yes Freeze+Listen is longer than Record+Listen.

Adq
1. Record+Listen -> Good -> Ready
2. Freeze -> Listen -> Good -> Ready 1 is faster.

 
From a theoretical standpoint, yes. But I have to wonder how many people here have actually tried to capture randomness as part of their work, and dealt with the reality of the situation.
 
(1) is faster only if you trust yourself that you were able to evaluate the part you played while you were recording, believe all the random changes were correct, and are sure you will not need to redo the track. Otherwise, you will need to re-record it again, hope you get your moves right, and evaluate. Every time you do this, you run the risk of not playing the part perfectly and having to start over again.
 
On the other hand with 2, once you get the part right, you can freeze and evaluate. Don't like it? Undo, freeze, evaluate. You can keep undoing and refreezing until you get a part that's just right, without having to play it over and over and over and over and over and over. 
 
I'll take freezing over real time recording any day, because it can generate possibility after possibility until one comes along that's perfect - and I only have to play the part right once. And of course, MIDI can record controller tweaking so there's absolutely zero reason I would need real-time recording for control tweaks.
 
As to real-time recording being a "professional" feature, one of Cakewalk's many truly professional features is how well it handles track sync, sample accuracy, and delay compensation. As pointed out by azslow3, you jeopardize that by unbalancing the system.
 
To me, recording VSTi audio in a way that forces you to re-do a part every time if there's one little aspect you don't like - whether it's the way you're playing or the effect of the randomness - is pointless. I would hate to see Cakewalk spend their development time on something that offers potentially faster operation in one very specific set of circumstances at the expense of jeopardizing timing stability.

And for those concerned about latency with a physical loopback, that's not an issue. You can monitor the signal before it goes out to the interface. Then you've captured your performance as a "safety," if nothing else. If it's great, keep it...done. If not, freeze, evaluate, undo until you hear random changes you like.
 
I just don't see how this is such a big deal, or even a relatively minor deal.

You didn't understand me right. I mean Record+Listen is faster when you have midi recorded already, and want to find sweat random take. Now you need Freeze then listen, and it can be 10 or more times, and Freeze could take considerable time if you use huge effects set on the synth. With recording possible you just start recording (you do nothing, just listen), if you don't like it, you just stop and start recording again, if you like it you are done.
It is more than common practice in electronic music when you use glitch synths. I use it always, in every project, and it is not something weird, but very usual electronic music. And in other post I have described situation when you can't achieve best result without possibility to record synths' audio.
I understand that it will never be implemented in Sonar, but the fact is that in routing field Sonar is worse than almost any other DAW, just because of functionality lack. It doesn't prevent me to love Sonar very much, but it is the fact.

Adq
It is more than common practice in electronic music when you use glitch synths.

Can you provide for me an example of such synth?

Adq
You didn't understand me right. I mean Record+Listen is faster when you have midi recorded already, and want to find sweat random take. Now you need Freeze then listen, and it can be 10 or more times, and Freeze could take considerable time if you use huge effects set on the synth. With recording possible you just start recording (you do nothing, just listen), if you don't like it, you just stop and start recording again, if you like it you are done.

Okay, I didn't realize you assumed a part had already been recorded. I thought you were talking about true real-time improvisational recording, where you were playing and listening to a random overlay. 

Adq has hit the nail on the head. Where this feature comes into its own is when there are random elements at play and the fact that you may react to those random elements and as a result of hearing something random you do something else as a result.
 
None of that can happen with bounicing obviously. Recording midi is also not the answer as well. I have got some VST's that never repeat themselves even when the exact same midi data being presented to them.
 
Studio One has it and it is a very cool feature. Also they have set it up so that a feedback loop never results as well.
 
I can create the feedback loop externally and use the latency of the system to create tight flanging and echo effects as part of the sound design process but you can do that anyway with external patching. That feedback effect when controlled can also be very cool.

Jeff Evans
Adq has hit the nail on the head. Where this feature comes into its own is when there are random elements at play and the fact that you may react to those random elements and as a result of hearing something random you do something else as a result.

But he clarified the use case I was addressing by saying the part was already recorded ("I mean Record+Listen is faster when you have midi recorded already, and want to find sweat random take"), so he's not going to be interacting with the part, only processing an existing part. 
 
For what you're describing (and a different use case he described), I could understand the concern if it wasn't possible to achieve the intended goal with SONAR, but you can. It just takes either a few clicks for software loopback, or two patch cords and inserting a send for an external physical loopback. That will hold me over until a 64-bit version of Jack appears (of course there are zillion ways to do loopback if you run WDM, but I prefer ASIO).

I asked for this years ago... Some synths have random elements - such as oscillators that don't start with the start of the midi stream. When you play these synths, you can often get really cool sounds and rhythms going by "playing the randomness". With these synths, it NEVER sounds the same when you play the midi back through the synth. 
 
This is one reason I do my creative work in Ableton - it has more flexible routing. 
 
Say what you want about freezing - the point is this is a NEEDED feature for those of us that want to capture that creative performance by "playing" the randomness of a synth.