Cisco WAN :: BGP Peering Causes 7200 To Crash?

I have two 7204VXR with NPE-G2 and 1Gb of ram. One router has 2 eBGP peers and the other has 3. The routers receive all internet routes from the 5 peers and send 2 internal routes. There is an iBGP peering between both routers. On all peers I have a route-map to send only our routes.
 
All was working fine since a couple of months when I suddenly saw an increase of memory on one of the router (router B), 1 hour later the memory was 100% and router crash and reboot. The other router (router A) with the same hardware capacity, same ram and same amount of routes was working good. After router B restart, I shut all eBGP peering on it, keeping only iBGP with router A, ram used was the same as router A (about 50% used) but CPU was about 30% used by process Router BGP whereas router A which has active traffic and active eBGP is only 20% and bgp process i almost 0%. Restarting peers one by one on router B cause the same issue, increase of memory then crash, even with only one peer.
 
What I suspected :

- A peer on router B but I can't isolate one because the problem appears with each taken one by one

- Not enough memory,  but router A has the same number of routes and don't have any problem

- IOS version ? same on both 12.4.(15)T1

- Why process router BGP use 30% on router B when all eBGP peers are shut except iBGP and no traffic pass.

- A routing loop but I only send internal routes to peers and only have one iBGP session with no sync nor redistribution with an IGP
 
Of course I can't run any debu ip bgp on routers as the number of routes is very large (300K).