Consider this ...
 
A Lesson on Buffer Size - Berklee
https://www.youtube.com/watch?v=5m8GoJeqras

A couple other things to consider since I am at a computer and can see the sig now.
 
1) With significant resources available, not only do programmers get sloppy, but the end user does too. Piling on CPU intensive items (VSTis, plugins, etc.) will chew up available resources quickly. Bear in mind that an audio track running through 18 FX requires each and every sample to be processed 18 times before it hits the buss. Freezing tracks of this type, or out-right baking them (bouncing to new tracks, then archiving the original), will let the CPU "breathe" and have more time to think while filling the buffer. Lowering the buffer to extremes in that situation makes it harder for the CPU to keep up, so being aware of what VSTs you are using is important. Trying to monitor a VST with look ahead can turn into a nightmare that even a smaller buffer won't solve.
 
2) I also have the i7-2600K and do not run mine overclocked. I have OC'd it a few times to check it, and in that situation the heat begins to throw faults from the CPU. Electrical components begin to suffer dielectric breakdown when exposed to excessive heat, and upping voltage just creates unnecessary heat at the CPU core. Even in fairly intensive projects, my CPU may get to 20%(ish), and the only things that will peg it are video rendering and encryption functions. Just to see this for yourself, PassMark has a benchmark utility that you can run and check different levels of OCing. Some levels of OCing are transparent (like 5% or so IIRC), but as that voltage goes above this, the faults start to come in, and each fault is going to find its way into the buffer at such a low buffer value - the CPU doesn't have time to calculate it a second time.
 
A couple things to consider is to back off the OCing in addition to upping the buffer to 96 or 128. After I saw your sig, I am not convinced that the OCing is not also contributing to what you are experiencing.

Hmmmmm......I can try backing off the OC. It has been running cool for over a 3 years. I have a nice case and good fans with an excellen passive cooler do I believe I am in good shape there.
I don't want to up the sample buffer unless I have to; that is the point of the thread!
Thanks for the well thought out response.

I missed the "32-sample ASIO buffer" size when first responding.
 
The problem is trying to run heavy loads at a 32-sample ASIO buffer size.  
With buffer sizes that low (.7ms at 44.1k), there's no margin for anything less than ideal circumstances.
 
Low/consistent DPC Latency (not just acceptable) is a must
 
BFD-3 is a wonderful drum plugin, but it's disk-streaming isn't the most efficient (compared to Kontakt).
If you've got a fast ride cymbal passage (with lots of ghost notes), polyphony can pile up fast.
With several disk-streaming sample libraries, you could easily exceed the capability of a 7200RPM "Samples" HD.
First thing I'd do is get BFD-3 library on SSD.
 
The MR816 is a nice audio interface, but it uses a large hidden safety-buffer.
With a smaller safety-buffer, you'd be able to (more comfortably) run it at a 64-sample ASIO buffer size.
For running heavy loads at a 32-sample ASIO buffer size, you'd do better with a PCIe audio interface (RME, MOTU, Lynx).
 
A recent make (well configured) machine would allow running heavier loads at a 32-sample ASIO buffer size.
 
 
I don't want to discourage the OP.  Effectively working at a 32-sample ASIO buffer size is demanding.
The solution isn't simple... as it involves every facet.
Any weak link throws a monkey-wrench
 
 
 

Thanks Jim,
Next up is an SSD for the sample drive. I want to upgrade all peripherals before migrating to a new PC. Dropping $2000 on s new interface is not in the cards for a while. Actually, the MR816 really is a nice unit. It actually works better under Windows 10 much to my surprise. If my only problem is a 32 sample buffer I can live with that! I sure that a 1tb SSD would drop below $200.
Anyway, you aren't discouraging me, I think that my machine is running crazyingly well. The machine is around 4 years old or so and my biggest concern is that the 350W power supply might crap out.

Mettelus:
Took your advice and monitored my CPU usage and dropped the OC to 4 GHZ. May drop it further but it is running with a peak temp of 54c and an average of 36c with the PassMark tests running.
Again, thanks

scottfa
...
Anyway, you aren't discouraging me, I think that my machine is running crazyingly well. The machine is around 4 years old or so and my biggest concern is that the 350W power supply might crap out.

Right you are to be concerned. I have a 400W supply, and it was just enough to handle the new Gigabyte GeForce video card I used to replace my old ATI Radeon that wasn't going to get a Win10 driver.
 
If you're going to add a video card, verify its power requirements. Also make sure that your power supply has the spare cable with the proper connector to power the video adapter you're considering.

Just to wrap this up. After researching graphic cards I found little 2d improvenents. There may be something to gain by having a new card not take up system memory. Anyway, after really looking at the Task Manager I lowered my OC to 4 GHZ. In the bios I upped the internal graphics card frequency by a little.
In Bfd3 I increased the amount of RAM used. This helped a lot.
In Sonar using the "e" key to bypass all effects helped. Also using the pause key helped a little.
The end result is recording with a 32 sample buffer is better. In most cases I will try this out. I am very confident that a 64 sample buffer is rock steady. I have measured the RTL and at 64 sample buffer it is around 7.4 ms. A little high, but doable. Resplendence crashes using the 32 sample buffer unfortunately.
Anyway, thanks for all the great info.