Performance issue

[absolutely_free at libero.it]

Hi,
Since few days I noticed very high load on my mailserver (Centos 6.6 64bit, 8 GB RAM, 2 x  CPU 3.00GHz 
I am using Dovecot + Postfix + Roundcube + Nginx.

I have about 10000 users.
Spool is on network attached storage (Coraid).

File system is ext4 (mounted with noatime).

Problem appears almost every morning (while load is normal during afternoon).

I suspect that this can be related to some user that have so many messages in his mailbox.
How can I troubleshoot this?

Here some messages that I got in maillog:

Warning: Maildir: Scanning /var/spool/pop/domains/xxxx.it/YYYYYYYYY/Maildir/new took 71 seconds (1 readdir()s, 1 rename()s to cur/)

Warning: Maildir /var/spool/pop/domains/zzzz/WWWWWWWWWWWW/Maildir/.Trash: Synchronization took 74 seconds (5 new msgs, 0 flag change attempts, 0 expunge attempts)

dovecot: imap(xxxx at zzzz.it): Warning: Inotify instance limit for user 89 (UID postfix) exceeded, disabling. Increase /proc/sys/fs/inotify/max_user_instances


tail: inotify cannot be used, reverting to polling: Too many open files

My relevant dovecot conf:

mail_location = maildir:/coraid-s2l2/domains
namespace {
  type = private
  separator = .
  prefix = INBOX.
  inbox = yes
}
mail_uid = 89
mail_gid = 89
mail_fsync = never
first_valid_uid = 89
first_valid_gid = 89
maildir_very_dirty_syncs = yes
mbox_write_locks = fcntl



thank you very much!