Toggle your prefer_direct and/or nonhierarchical_direct settings and look for any changes in the access.log.
Enable netdb and view the netdb cache manager page after Squid has been running for a while.
If using ICP or HTCP, count the percentage of requests that experienced a timeout waiting for replies to arrive.
If you used enable-cache-digests and have a reasonably full cache, disable the digest_generation directive and note any change in memory usage.
Use your operating system's packet filters to block ICP or HTCP messages to your neighbors. How quickly does Squid change their state from alive to dead, and back again?