Recording with more than 44.1khz and 16Bit

Recording in 24 bit mode is definitely much better.  The digital noise floor drops way down.  Allowing you to track and mix at a lower overall rms level. eg around -20 db for example.  That way you will have 20 dB headroom above and still well over 100 dB of signal to noise below.
 
Dithering does not introduce artifacts.  I have yet to hear any artifacts in dithering.  In fact it is adding in noise and making things sound better.  I stay in 24 bit all the way right to the end and use my PSP Xenon limiter to do the dithering at the very last minute.  It does it very nicely and sounds great.
 
Bounced wave files for me do not sound any different than listening to the session in real time either.  I am not on Sonar though but on Studio One but I never hear any difference.  People have said this before too but I don't agree with it.  There must be something else going on for that to happen.
 
Higher sampling rates during recording is another question though.  From my experience it does not seem to do a whole lot.  However rendering some virtual synths at 96K can sound different.  You guys with Sonar now have the option to up sample in those sensitive areas which is great.  But the 24 bit depth will give you a better sound though. I often record at 44.1K and 24 bit if it just a straight ahead audio session only.  That way I only have to alter the bit depth for any CD burning etc..

bewerber2
 
But the point is that I always have the feeling that my bounced *.wav-files sound different than my DAW playback. That's why I asked this question. Since I am big friend of WYSIWYG, I am always frustrated when I export a project to a *.wav; 

It might help to think of it in terms of photography. You take a 10 megapixel picture of something and then turn around and take the same exact picture of the same exact subject with a 20 megapixel camera. To the naked eye, both unedited photos will look more or less exactly the same. The difference comes when you start editing and manipulating. Even though both photos look essentially the same, the 20 megapixel version contains much more information about what original subject actually looked like. Therefore when you start editing and applying processing, you end up with a much better result from the 20 megapixel source than you do from the 10 megapixel version. Even if your end result is going to be 10 megapixels. 

It is exactly the same with audio. You will be hard pressed to tell the difference between an unedited 16 bit recording and an unedited 24 bit recording of the same source. Unedited, both will sound exactly the same to all but the most talented ears (think like 10 people on the planet could tell the difference maybe). 
 
But once you start applying processing, the more information your source contains (i.e. bits), the better your processors will be able to do their job accurately. 

That being said, here's a way to consider it that is much more literal. The nature of audio and bit rates is such that 16 bit audio only actually uses all 16 bits when the audio level hits 0dB. If your signal is less than 0dB, then your audio is playing back at something less than a true 16 bits. 

And since nasty god awful things happen if we exceed 0dB in the digital world, it stands to reason that most of our audio is going to end up being something less than 16 bits if we start with a 16 bit source and then mix so only the highest peaks approach 0dB. 


However if we start with a 24 bit source, we've got lots of headroom to process and mix our output to something less than 0dB and still end up with a result that exceeds 16 bits. This then allows us to dither the end result down to 16 bits without losing any detail that would be detectable to most ears. 


As for your belief that your 24 bit audio always sounds different after being truncated down to 16 bits, I can only speculate. Are you listening to the resulting 16 bit output on the exact same system with all of the exact same processing in the signal path? Or are you creating the 16 bit output and then listening to it on other systems?
 
And if the system and processing in between is exactly the same (i.e. same speakers and same everything between the software and the speakers), have done any true blind a/b listening tests where you try to identify which version you're listening to without otherwise knowing? If not, I would strongly recommend you do so. You may be shocked at how much our preconceived notions impact what we think we're hearing out of the speakers.