Just a quick note about Firewire and USB. (it'll take longer to digest the rest of your post).
Firewire uses an 8KHz framerate "heartbeat" -- which means with really clever hardware you could probably get under 500 microseconds latency (I haven't seen it done, but it's theoretically possible). There's a major tradeoff between audio buffer size and bus utilization, of course.
USB uses a 1KHz framerate -- 1 msec per frame. Plus, transfers have to percolate through the driver stack; they're not handled in HW as with Firewire. Big difference, as far as low latency performance is concerned.
Of course, Firewire bus speeds are steadily increasing (3.2Gbps in the labs currently, I think). I'm not aware of any work to speed up USB 2.0 (especially since the development of USB 2.0 was largely motivated by Apple's appalling decision to charge per-port royalties after the industry committed to Firewire for high-speed peripherals -- a decision that Apple later backed away from, but which seriously messed things up IMHO).
I also think the basic Firewire isochronous transport protocol is much more sensibly designed than USB, and Firewire media transport protocols far more sensibly designed as a whole -- but that's a different issue.
Which is not to say that PCI and PCIe don't offer more bandwidth, and substantially lower latencies -- of course they do.
People seem generally pleased with latency and other performance characteristics of the RME Fireface, but I haven't tried one myself. Have you found it to be limiting in practice (in terms of track count, realtime latency, other performance characteristics) ?