I don't think people who really know what they're talking about say that. Sure the JVM is slow to start so there's latency involved in getting Java up and going, but Java is very fast over the long term.
Java also eats a lot of RAM in many applications (J2EE apps can easily eat a few GB of RAM), so combined with the multi-second latency for startup, you're basically confined to a single programming model: One giant, long-running, multithreaded program.
We recently went live with a suite of applications (5 apps all together, both web and backend), processing tons of data in "real-time" in multiple threads and after a few days of usage someone noticed it was running with default 512M of RAM (8gb planned).
Considering the load it is processing - it is very impressive. And yes, it is using all of J2EE in WebSphere with mq, hibernate, spring core, spring integration, etc.
BUT I could not make JIRA work with 512 linode, had to upgrade to 1m and later on switch to jira-cloud as it was failing all the time.