-XX:CMSWaitDuration=0 -XX:+ExplicitGCInvokesConcurrent can cause a deadlock because of a race between the setting of the request flag and the CMS thread going into a wait() without checking the disposition of the
flag. CMSWaitDuration=0 is a potential (interim) workaround of sorts for (some instances of) 6412868.
Small change, probably worth doing sooner rather than later, especially given its relationship wih 6412968.
When fixed, this will also have the beneficial effect of (in some configurations) reducing the variance
in the latency of concurrent full gc's seen at the caller.