I don't know enough about how Freenas is using the memory to determine what I will risk if I go about 32TB , system stability or just less cache ? What you guys can advice me ?
You'll probably be fine.
The further up in size you go, the more flexibility there is in the rule-of-thumb. It is more to prevent people from doing catastrophic stupid like "why does my my 100TB pool on 8GB not work". That isn't going to work out and I would expect it to be potentially unstable. I deliberately rewrote the handbook section on memory sizing to be a bit vague because of this.
For a busy departmental fileserver with a small amount of storage, you almost certainly need more than 1GB per TB, at least if you want good performance.
But for light duty file service, which is generally what archival or video storage is, with (relatively) lots of RAM, you can go lighter on the RAM. You might lose some performance, especially write performance, since the task of writing to the pool involves having lots of information about the pool in-core, such as where the free blocks are. Keeping more free space available helps to mitigate this.
As long as you're past the 8GB mark and you're not stupidly outside the rule-of-thumb size (maybe more than a factor of 2x-3x) I can't think of anyone who's had stability issues, just performance stuff like Cyberjock talks about. Scrubs take a backseat to other pool traffic anyways so I'm not really all that shocked at his two days, or even if it were to be a week. The 30TB (40TB raw) filer I talked about above is at 42% capacity and did its last scrub in 20h13m, which implies it was doing around 550MBytes/sec from the pool average. As capacity goes up, scrub times are just going to be longer.
