Another question would be if there were differences between Intel and AMD or different CPU types and RAM.
A few years ago I have already read that AMD renders reverb different to Intel (AMD needs more CPU power). And there maybe differences how the CPUs are processing and rendering the audio signal.
 
 

maybe changes to the DPE between versions?

I'm glad my little philosophical musing on subjectivity 'struck a chord' with a few people, it's rather an important topic for me - I really do think that a misunderstanding of confirmation bias is at the heart of much human misunderstanding, conflict and misery - as well as making us wonder whether we can trust our ears. Luckily, the scientific method exists, pretty much entirely for the purpose of overcoming it. Unluckily, most people don't understand that very well either, and an alarming number appear to have now decided that science (and, you know, reality-based, fact-checked journalism) is somehow the enemy because it sometimes tells us things we don't want to hear, or things we've decided to believe otherwise about. 
 
For the OP, here's what I would suggest: Find someone to help you do a real, double-blind A/B test, where you try to tell the difference between the two - whether rendered files or played back in the program, up to you - based on what you think you're hearing, do it a whole bunch of times in a row, and see whether you're right a statistically significant percentage of the time. It's not really difficult to do, but you have to be determined to find the real answer, not just confirm what you want to believe. Most people are actually not committed enough, and/or actually *don't want to know*, because they are afraid of the answer and would prefer to continue to believe what they already want to believe.
 
If so, that's OK, but you have to accept that you *could* know the real answer but have chosen not to. With regards to the difference or non-difference between two long-outdated versions of Sonar, the stakes are pretty low, nobody gets hurt. But it's an interesting exercise in confronting confirmation bias and attachment to beliefs in general, and remember: the stakes are much, much higher in other cases. And that's all I'm going to say about that, as this is not a political forum!

From the first post, one can think there can be no difference, but... there are still several places to check:
a) is that "dead silence" also dead on the level monitor or just sonic? Sometimes it is possible to hear the (objectively existing) difference in 2 files when you listen separately but not in the phase inversion. The "difference" can be just too small as the absolute value for the monitoring system. F.e. 16bit vs 24bit is on the level of -96dB. But "level monitor" still show it
b) related to (a), have you exported 24bit from both? It the project where you compare is also 24bit?
c) if (a) and (b) can not be an issue, both Sonars RENDER equivalently. But that does not mean the real time audio output is the same, so:
c.1) listen rendered file as the only track in each version, without any effects. Do you still perceive the difference?
c.2) if yes, precisely check audio driver settings in both versions.

Even the slightest difference in playback volume skews the results. Ditto if you're listening over speakers, if your head is even a fraction of an inch different when you listen to the two files.
 
The null tests indicate there is no difference with respect to SONAR, so if you really believe there's a difference, as azslow3 says - you need to look elsewhere for the reason.

But, folks, do a search on "improved audio engine in Cakewalk Sonar".  See what you get, and if it's more than just efficiency in processing, let's talk about it.  

If dithering is being done, even without bit-depth reduction, that has some random elements in it which could account for small differences. Though they'd be very minor indeed to put it mildly.

If two audio files null to absolute silence that means they're the same. Though the hardware - interface and monitors - they're being played back through might not be reproducing every run-through in exactly the same way. And as Craig says, move your head slightly between playbacks and you won't be hearing quite the same thing every time anyway.

Why don't you set up a blind (or deaf) listening test.   Make two audio files, one from each version.  And put the files in a playlist on some audio software, like Windows Media Player, and set it for "shuffle" play.   See if you can pick out which one is from which version.  If you can make the right choice, say 20 or 30 times, then you may be on to something.  But if it's like you suspect that it's "all in your mind", they your results will be like guessing which side will land in a coin toss.    Just a thought.

Thanks for all the replies so far.  To answer some of the questions:
 
As to format, the project is 24 bit/44.1K and was rendered to 32 bit WAV.

WRT any theory about me maybe wanting to subconsciously hear a preference it is actually the opposite. I was hoping that S5 would sound equal to S6 so I could simply stay with S5 where most of my projects are. So if anything, I was hoping S6 would sound the same or worse, but instead it seems to sound better.
 
BTW I did experiment moving my head around a bit in the sound field while listening back real-time and it doesn't account for the type of differences I'm hearing.
 
I was pretty meticulous in making sure everything was set the same between versions but I'll go back and triple check things.

Sonic

yeah you need to do the null test described above to really know if there is a difference, did you do that yet?