We came across this little issue in using the node stats API, where the total CPU time values are reported as negative 1 while the user and kernel CPU times are non-zero. The command used was:

curl -XGET http://:9200/_cluster/nodes/stats
...
"process": {
"timestamp": 1314943670627,
"open_file_descriptors": 112,
"cpu": {
"percent": 1,
"sys": "343 milliseconds",
"sys_in_millis": 343,
"user": "3 seconds and 226 milliseconds",
"user_in_millis": 3226,
"total": "-1 milliseconds",
"total_in_millis": -1
},

This is on the 0.18 series of elasticsearch, running on a Mac OS X 10.6.8 machine with multiple CPUs.

cheers.