ExtremeCap U3 works on USB 3.0 only: it's not sending a compressed mp4 stream, but it's sending each frame uncompressed. That's a lot of data. And I don't care the latency, since anyway I'm configuring the delays for a perfect (below the frame) sync.
That one has a latency of just 233ms, which is not that much: it's a uncompressed capture, frame by frame ( = if it skips a frame it won't unsync, it just skips a frame).
I was using a USB 2.0 capture device also, the Elgato Game Capture HD; that one (for the lowest USB2 bandwidth) was compressing to mp4, and had a HUGE latency (1533ms). But again, it did not matter that much: with the right delays on each capture device & each sound source, it was just perfect. Except the Elgato wasn't uncompressed frame by frame but it was a mp4 stream, meaning: if the USB isn't responding for a short time or the device skips a frame, everything gets unsynced. It's with those USB2 capture devices that your stream gets unsynced other time (each time a frame is lost you get an additionnal 16ms delay to your video feed that is adding up other time).
USB3 devices are good, if it's uncompressed, and of course it's stable (not crashing). You can freeze, skip frames, whatever, it's always frame by frame, so it always sync again when it's working back.
The final stream & video had all synced below 10ms (both video feeds & both sound sources): checked frame by frame with Premiere Pro.
But anyways, that's not the topic, hehe. That worked really well on that same exact PC, but on Windows 7, until I upgraded it to Windows 10.
I just bought two PCIexpress capture card, so that's solved, anyway (and delays will be much easier to configure with the two same cards). If that works.
(took two Elgato HD 60 Pro; they provide a beta driver that can deal with multiple cards)
I liked that ExtremeCap U3 because it has a special "feature": it was delivering 60 fps no matter the real HDMI signal and adds duplicate frames if needed. My game & game capture was pure 60.0 Hz (RGB HDMI)... but my cam is 59.94 Hz (NTSC HDMI). With the ExtremeCap U3, that "feature" allowed to add duplicate frames and have a perfect sync with the other capture device at 60 fps. I just duplicate a frame every 1001 frames, so it's not really noticable (that's just one duplicate frame every 17 seconds, at 60im/s you don't see it). I hope OBS will be able to do it also by software with my new Elgato PCIe cards (one at 59.94 the other at 60). But that's another (long) story.
Again I don't care the latency: it's important for those casting & gaming from a single PC, because you can't delay the local screen capture, and it's the camera that has a big latency (so you would need to be able to delay the screen capture to sync it with the cam). But with a two PC setup (or single PC but two capture devices), you can delay any capture (the camera, the game capture, the sound sources), so you can sync everything perfectly. Not easy, I admit, but once you know how to check the latency, it's quite straight forward (I just check the video frames sync & audio latency with Premiere Pro on a recorded footage and add the corresponding delay to that device).
(go check twitch.tv/zetmor this Sunday if you want to check it! hehe... I'm playing Rocksmith, learning guitar with it since this monday)
problem solved!