Do you have any debug output from syslog whenever these things happen?
Are all switches running the same EIGRP AS (I assume yes)?
Do both default routes get removed from the routing table or just one?
Are you using the standard K-values for metric calculation in EIGRP?
Just some rudimentary questions without trying to be too specific. Others will probably chime in. Usually it takes just one person to start the flood.
sstandard k values are used, both routes fall out of the routing table but remain in the topology table. All routers are using the same AS number. There are no syslog messages logged when it happens. This is happening on older 3560d and 3750e switches. On the 3750s I added and EEM script that tracks an Ip SLA ping to the Internet. When it fails the script does a soft reset of the neighbors and fixes it. For the 3560s we have to log in and manually reset them.