robert_e_bone
I am sidestepping the whole discussion - I am quite happy with the results I get with using 48 k, and I never have to worry about dropouts or otherwise taxing my system, in any way.

And that's as it should be. Unfortunately, I am a music and sound addict. I am always looking for a stronger high. 

Anderton
The plot thickens...
 
Actually, both Bitflpper and I might be wrong about 96 making a difference only with signals inside the computer. No less an authority than James A. Moorer wrote a paper that proposed, among other things, that hearing involves not just frequency and amplitude, but time and how it relates to localization when listening with both ears. He claims that most people can distinguish a time delay of 15 microseconds or more when a pulse is put into each ear, and that some people can differentiate delays as low as 3 to 5 microseconds. Given that a sample at 48kHz is about 21 microseconds and 10.5 microseconds at 96kHz, that means the minimum time delay most people can differentiate is actually less than one sample at 48kHz, but more than one sample at 96kHz. 

 
The sample time does not remotely equal the time resolution. This is a common misunderstanding of how sampling works. 
 
It depends on bit depth as well as sample rate, but the short answer is 48kHz has a timing resolution of FAR greater than 1/48,000.

Anderton
One more question: Wouldn't a higher sample rate also spread out quantization noise over a wider bandwidth? I also wonder about jitter. Wouldn't a higher sample rate distribute any jitter over a larger number of samples, which when interpolated and filtered, would give better results?

In terms of quantization error, given that internal processing is done using 32bit floating point (or higher) the short answer is that the sample rate is irrelevant. 
 
Jitter is really only an issue at the converters, and of course oversampling (often at a much lower bit depth) is commonly used in converters for a variety of reasons already.

Anderton
 
But the point of the comparison is about the accuracy with which the sound represents what the synth was generating. (If I had any synth sound in a track with that many highs, I'd run for the QuadCurve's lowpass filter.) 28 has high frequencies that simply don't exist in 27, and they're not distortion byproducts.

I am having some difficulty discerning how you are determining what the synth is actually producing and how you are deciding that one recording is more accurate to that production than another. More pleasant, maybe, but how accurate?  Many people find the distortions inherent in analog recording/processing to be "warmer" or whatever cork sniffing metaphor you prefer. But it is pretty easy to demonstrate that analogue recordings of a reference source are less accurate than a properly done digital version using instrumentation. With a signal produced in the box, what is the proper A/B comparison to determine if it is accurately recorded?

DrewFX and Bit, I think you're both awesome!
 
I've been using double rates for a while now. I tried Quad rates on a session but I'm sticking with double rates (96). My UCX may like it better, idk.
 
I'm not running into problems on my computer using double rates, so shrug... there you go.
 
Anderton, I think it's so awesome that you went straight for the evidence based choice.
 
DrewFX, what would the time resolution of 24 bit 48kHz be? Can you figure it out? I'm curious.
 
If this guy
http://en.wikipedia.org/wiki/James_A._Moorer
thinks it might matter, I'm curious to learn more.

Anderton
 

If Z3ta+ falls into that category, then I maintain that it's a design flaw.

 
So do you consider including oversampling compensation for a design flaw, or simply good practice? If the answer is simply good practice and therefore it must always be done, then that restricts how many instances you can use and balance out CPU consumption. In some cases, there will be no audible difference whether the synth is loafing along at normal speed or oversampling, and many might choose not to oversample as a tradeoff for something like being able to overdub a guitar part with lower latency.
 
So the bottom line is this: if oversampling and filtering are good practices to compensate for deficiencies caused by generating undesirable harmonics that can create problems with lower sampling rates, why is recording at 96kHz inherently negative if it too can compensate for deficiencies caused by generating undesirable harmonics that can create problems with lower sampling rates? Aren't they the same thing, except one is applied as a global preference as opposed to a localized one?

I would say that if a modern plugin is aliasing, then it should at least give the option of oversampling itself. It's sort of a no brainer.
 
For older plugins that were developed when CPU power was much more limited it's more understandable why they might lack this functionality (particularly if they were already pushing the CPU envelope at that time), and that's where your solution makes sense.
 
But the point is that you're doing something that the plugin's developers should have already done in the first place. And for recent plugins they probably already did do it. And if it's a recent plugin and they didn't oversample - even though it was clearly necessary - then I would say it is absolutely a design flaw.

I've been recording at 96/32.  This way I don't have to worry much about slight clipping as I go.  I can't see a drawback because disk space is pretty cheap now.  So I feel ya brother!  Bouncing down to 44/16 is pretty awesome too!

gswitzDrewFX, what would the time resolution of 24 bit 48kHz be? Can you figure it out? I'm curious.
 

Something on the order of 0.0000000000002 seconds.
 
Note that 1/48,000 = 0.0000208333333 seconds (Craig's 21 microseconds).

Anderton
This site is really interesting: 
 
http://src.infinitewave.ca/
 
Check out the difference in sample rate converters between Ableton Live 7 and Ableton Live 9, and that was only a few years' difference...then consider the days when we had 16-bit engines. Back then, conversions from 88.2 to 44.1 did sound better than 96 to 44.1. Fortunately that period didn't last long, but it did exist. 
 
Oh, and if you want to feel good about Sonar, while you're on that site compare it to a bunch of other DAWs. They used 8.5, but I assume the sample rate conversion didn't get any worse in the X-series.

 
I actually just submitted test files converted in X3 to that site a few days ago, but they haven't posted them yet. Opening them in iZotope RX3, which I believe is what they use for their "sweep" screenshots, they look better to me than the 8.5 ones. Less aliasing.

I was quite surprised at the difference between the two files. The 96kHz clip definitely has way more high frequency content, in fact the 44.1kHz sounded to me like it has some gentle low pass filtering in comparison. 
 
I have long noticed quite a large difference between having 2x oversampling turned on or off in Z3TA+2. It's not an open and shut case as to which sounds the best, however - it depends on the patch. For example when it's a bright plucky patch then the 2x oversampling sounds better to my ears. But for a warm sounding pad, I think it sounds better with the oversampling turned off - you get a much creamier "analog" sounding pad. 
 
As for Craig's two clips, I definitely prefer the sound of the 44.1kHz clip, it sure sounds a lot fatter and warmer, and the high frequencies in the 96kHz clip are way too harsh sounding. 
 
But don't people often say similar things about lower bit rates where some kinds of sounds are concerned? For example the much sought after "phat" sounding drums from an old 12-bit Akai sampler.