Check about GC activities : use verboseGC to generate some statistics and then study the % lost in GC.
check database connectivities and then study slow queries.
ensure the machine is not swapping.
verify system settings (kernel stuffs, patches, background processes, JVM version).
Even part of a cluster, each IS behave individually so a local activity, a network issue or a misconfiguration at kernel level may create issue on only one member w/o impacting others.