We have worked with both, but I would vote for the former - there are less points of failure and you don't need an IT person to make sure every node is running smoothly. We built the latter using multiple 4-channel MPEG-2 streaming cards + MIDI racks and used it for all sorts of events (the biggest was 360-degree back in 2006: http://www.motionlab.at/projects/content-production/urban-ar...) but there were always problems and I often had to rip hardware/software apart to figure out what was wrong. That's why I ended up writing a minimalistic tool for controlling the cards, the vendor's included software wasn't meant for our type of work - thankfully there was a decent C++ API available. Workflow-wise it's also much easier/faster to render one big video than a bunch of smaller videos, not to mention you can check final output on one screen. We wired up 8 mini-screens in the rack to preview the streams, but they were quite small, making it difficult to determine if the footage was ok.