I've already tested this, you can read it in message if you read it properly unfortunately it still doesn't work. I have to use watchdog=false because on systems with 8+ gpus you cannot run this miner properly, because the cpu usage in the beginning is too high - first 2-4 min that the miner needs to initialize it self to max speed,...then the cpu lowers to normal state 40-60% on low cost G3930. I've reported this bug already and few others too.
What you're experiencing is the (bad) nvidia driver allocating GPUs. That's not bminer's fault (you can start any gpu app and you'll see the same). That delay is particularly long in Linux, but I don't believe you when you say it's 2-4 minutes. I use G3930 systems too and the delay is 80-90 seconds with 13 GPUs.
Your other connectivity issues can't be the miner's fault. The pool or your own connection happened to have issues.
Regarding the other posts: I'm still a amused by the trust people put in tests done with a single card ...
I don't defend bminer - the developer is way too shady, sorry (see
here) - but some things are unrelated and can be explained.