Hello Mohamed,
please consider the following facts:
Saquid's wan links are not equal in performance and type see initial post.
Here, use of variance would be wise to make possible to take in account these differences.
My first impression was that the drops were only on the slower path, for the default equal cost paths so I've suggested to use bandwidth and variance to differentiate paths to be able to put more traffic on the better link.
However, later show commands from Saquid show that output drops are present in both tunnels and actually counter is greater on tunnel going on better path.
Saquid has already tried with a different IOS release with no improvement.
To be noted that Saquid says that drops happen only if both tunnels are up and he doesn't see drops if only one tunnel is up.
We need to be sure that tunnels are stable and that no recursion happens.
Recursion is : if tunnel1 destination is advertised inside EIGRP running on tunnel1 or tunnel2.
Saquid: look at logs to see if tunnels have up/downs events also check with both tunnel up
sh ip route t1-destination
sh ip route t2-destination
the destination have to be learned by different means then the EIGRP protocol running on the tunnels.
This is the requirement for stable tunnels and correct routing. Eventually EIGRP routing needs to be tuned (route filtering)
Hope to help
Giuseppe
