Hi everyone.

First of all the solution: Disable, Enable Rule Hit Logging, for all Rules.

Now a brief explanation:
With the help from Novell Support we found what was causing this problem and
found the solution.

The problem was that I have Enable Rule Hit Logging on most of my rules (I
though, if I want to have logging I need to enable them, sort like iChain,
one entry, one log). Wrong.
Enable Rule Hit Logging has nothing to do with Common or Extended log.
Enable Rule Hit Logging uses BTRIEVE, and BTRIEVE uses (is) NWMKDE. And
BTRIEVE isn't able to handle my load.
Putting in SYS:\SYSTEM\BTI.CFG, CacheSize=768000 like someone mention it on
this forum help a bit, but my server still stays 100% CPU0 for a long time
(2000 users, 6 Mb/s connection to the Internet).

Rule Hit Logging should be used only for troubleshooting of rules, otherwise
it should be disable.
This should be stated clear in the documentation and on BorderManager
Performance Tunning TID.