Okay, I've spent most of the day in SONAR, working on improved compatibility and performance with fairly dramatic results:

• I discovered a bug in the reViSiT audio engine that causes SONAR (and possibly other hosts) to crash with buffers over 1368 bytes. This bug was causing reViSiT to copy 4 times as much memory as required, at certain points. As a result, not only should the fixed version be more stable, but also better performing (as it will be copying less memory). Odds are your crashes were caused by this bug.
• I've removed some code in reViSiT that could potentially interfere with a host's attempt to use multi-processing. In earlier versions, this code gave reViSiT a slight performance boost, ensuring that reViSiT got to do all it's processing on one core, without interference. With the more widespread availability of multi-core processors, combined with other optimisations in the reViSiT audio engine; this tweak is no longer needed. Also, now that this restriction has been removed, a few parts of reViSiT have be parallelised, which might also lead to performance improvements.
• In the late beta versions, I used the Intel C++ Compiler rather than the Microsoft one, in order to improve performance. Although reViSiT did run a bit faster, the Intel compiler complicated the build and debug processes, and was subsequently abandoned. More recently, I have made changes to the audio engine that allow the Intel Compiler to take make even more optimisations, and thus I am once again investigate whether the performance increase now justifies its use.
• Larger audio buffer sizes appear to interfere with reViSiT's ability to keep time - notably, when to trigger frames. While this doesn't lead to pops, clicks or crashes, it does mean that reViSiT's patterns don't sound correct in such situations (e.g. effects and envelopes). The workaround is to reduce the host's latency (e.g. to less than 20ms) - but I'm investigating further, as to what exactly is going on, in the hope of coding a fix. This bug appears to exist in all hosts and drivers, but may have appeared more noticeable in SONAR, since the default buffer size is quite large.

As a result of all this, the version of reViSiT I am working on now, otherwise appears to be very stable and performing well, in SONAR  - working with Use Multiprocessing Engine enabled, in WDM/KS, MME and ASIO driver modes, with no pops or clicks, even with a latency of 1.0 msec (WDM/KS, 44 samples, 2 buffers). Naturally, mileage varies depending on your sound drivers. For example, these results were taken with the newest drivers (v1.1.1) for my Mackie Onyx Satellite - previously, this soundcard barely worked with SONAR.

So, if you can just hang in there for a week or two, I hope to produce a new version of reViSiT that should address the issues you're facing.

All the best,
Chris