Half the connections to one VLAN fail behind a summarizing distribution pair

The building distribution pair DSW1 and DSW2 runs EIGRP 100 to CORE1, and each advertises only the summary 10.20.0.0/16 on its uplink (ip summary-address eigrp 100 10.20.0.0 255.255.0.0). Each has a static default route to CORE1. Every access switch has a trunk to both that carries only its own VLANs, and HSRP gives each VLAN a gateway. There is no link between DSW1 and DSW2.

This morning the optic on DSW1 toward ASW3 failed. ASW3 carries only VLAN 30 (10.20.30.0/24). VLAN 30 users still have a gateway through DSW2, but they report that about half of their connections to the data center behind CORE1 time out, while the rest work fine. Other VLANs are unaffected.