depending on your aspect ratio of the table you could go for stitching feeds from mulktiple cams… but the BIG question here is that u’ll need shit loads of processing power to process feed from 8 cams in real time… for sure one pc cant handle it (to add to that i am not sure if a single pc can connect to 8 cams at a time… well it shud i guess)… even if u’re using 640 X 480 res (.3 MP).... u’re talking bout processing 2.4 MP feed at atleast 20-25 fps (for smoother feedback) in realtime…
i have been using a octa-core (quad core dual processor 2.0 ghz) with 4 gig of ram… and tried processing feeds from 2 sony handycams (1394 connectivity 720X576 res ... comes to .45 MP) (this was not for multitouch but chroma and 3d vision) and was just bout able to get a output at 30 fps post processing ... the processing involved here was probably a little more that what is required for blob detection. but my point here is that 8 cams could be a problem to process…
i would reccommend what WILL said.... try a wide angle lens on a firewire cam that gives you a good res feed at a very hi FPS. but that depends on the aspect ratio of your surface.
another suggestion would be to have cams connected to different pc’s and you can listen to touch events through network but then u’ll need a seperate engine runnin on the main comp that maps the X,Y co-ordiantes from comp no. 1 - 8 to a local co-ordinate system (for eg: if your output res is 1024X768 , then the 0,0 from the second comp will be equal to 129,0 on the local co-ordinate to which your app should respond) ... i am suggestin just a logic here if someone can work on the programming part at your end....