Nicholas,
After some installing and uninstalling and re-installing... I was able to deduce that the newer versions of the ATEM software (6.4/6.5) seem to conflict with my USBtoSerial adapter. Since using the newer versions did not seem to impact my original problem one way or another, I decided to drop back to my originally installed version of 6.0 (with Desktop Video 10.1.4).
I then spent some time combing through my code and experimenting looking for details that might give me some idea of why I could randomly get the BMDStreamingServer.exe to start leaking memory. I finally had a eureka moment and am now able to replicate the issue on demand. Even better, I was able to isolate the issue is external to my code, as I can get the same thing to happen using the ATEM Software Control.
I prepared two short youtube videos that demonstrate how to get the ATEM software control to trigger the BMDStreamingServer.exe to start leaking. Basically, the short version is this: if the capture is stopped during a period of high CPU utilization, the leak will exhibit itself.
My development environment may be more susceptible to this problem as I am running within a Windows 8.1 virtual machine on my MacBook Pro. The VM (hosted by Fusion 6) is an i7 but only a single core. I'm not sure if Intel ever manufactured a single core i7 (I kind of doubt it), so it may be the developers have made some threading assumptions that should be OK in the 'real world' but do not hold up when virtualizing. Unfortunately, I do not currently have physical hardware available to test this on real iron.
To see a demonstration of the leak and how to replicate it, visit:
Thanks for you help!
PS.
I will also put the above into an email for developer support as you suggested.