i have only 12Go (6 per CPU) , do you think it might be a problem ?

Half a Gig per active CEP2 thread... it can be done. The odds are that all 24 wont top out at their memory needs at the same time, but if they would, you'd be seeing some memory swapping (and SSD ware), and those are already 1Gb+ at times for the CEP2 jobs. It's the reason that WCG specified a per-task minimum requirement of 1GB RAM and 2GB of disk space... so that volunteer machines wont come crashing down.ù

With 99% efficiency and task validating you've got a "Diablo Countach" at your hands... your only problem the system clock, if that is a problem if it can be Synched hourly with the web or another system on your LAN. Notably, BOINC does not like clocks being set backward and forward. Depending on client version you'd see the revert to previous checkpoint as the client might think there's a timekeeping problem. With the 7.0.28 [test] version I've found that to be more tolerant [set the date back a few months, wait a minute or 2, then forward to present, without stopping the client will reset some time counters and deferrals], but have not tested this with CEP2. With 24 jobs going back several hours is from an efficiency point of view costly, even when the client reports 99%. Your client message log would tell if jobs got restarted with such clock/date yoyoing.