Heya Superroach.
I got some really nice test data for you to work with, if you want.
First, to explain the setup that I'm using, I'm using XSplit in tandem with another program called DXTory. Basically, what DXTory is is an optimized version of the gamesource feature you guys are trying to implement right now. The nice thing about it is that it is very efficient, and takes far less resources than both XSplit's screen cap and game source cap.
How it gets used in tandem with XSplit is that it records the screen data to a virtual camera, so XSplit treats it like any standard webcam source, so you just plug it in, and there you go.
Anyhow, I was playing some SNES games on ZSNES. ZSNES is very light on processor load, and the capture wasn't a full 1080p window, so that also makes CPU load lighter. I was casting at 720p, quality setting 10, 55 FPS, and I was receiving results consistent to the video linked in my original post.
Checking the bitrate, the highest I really saw it go is 2500. I tried XSplit bitrates inbetween 3500-4500 and no tweaking of that setting helped.
So, here is the test data!
1.
http://www.justin.tv/da_sud/b/292507471 - starts slightly jittery, but you can see right off the bat when it's going at full 55 fps. It then degrades into really choppy gameplay. Note that none of the chop was visible in game, implying the bottleneck here isn't my CPU.
2.
http://www.justin.tv/da_sud/b/292525632 - title screen there is full 55 fps. Then a few seconds in it gets jittery, with patches here and there of full 55 fps.
As I said, the highest bitrate I saw in Xsplit's red text was 2500, meaning there should still be room for 1000-1500 more, so I guess internet would be the most likely, however I've not really experienced issues with 30 fps, and internet diagnostics haven't really shown any issues as far as packet loss or latency goes.
Any insight, or do you need any kind of info from me? Thanks!