"CAMIF" = Camera Interface.
Thoughts only - I have no 1st hand experience of this.
If you can frame sync the sensors there should be no need to lose frames, changing sources during the frame sync signal.
The excessively enthused could add a hardware buffer to clock in frame data from the next source and then swap to that and load from it when current frame is finished. IF you can load fast enough to pull ahead before frame end you can then swap to real time and assign buffer to next in line. Two x RP sounds easier :-).
If you cannot frame sync then progressive startup of sensors so next in line is available soon after end of current may allow minimal loss. Below I show sensors starting 1.xx after the prior one but of course just staggering startups by say 0.1 of a frame time would mean that the next sensor is available soon after the current is finished. They will drift relative to each other - slowly so if under crystal control. Just resync when delay gets too long or short.
11111111 = frame 1 etc
111111112222222233333333444444445555555566666666...
.........111111112222222233333333444444445555555566666666...
...................11111111222222223333333344444444 .........
If sensors are fully asynchronous you could just look for next frame sync after end of current frame and switch to that. Order will (almost always) be non sequential but you should cover all in time.
Oh Wow!!! department:
Samsung's "multimedia" S3C6410X RISC Microprocessor offering.
1370 page datasheet :-)
I can't find it for sale :-(
424 pin FBGA - 13mm x 13mm
S3C6410X RISC Microprocessor
http://www.rasterman.com/files/man-6410-1.2.pdf