This is a little more accurate (since it counts actual reloads, not potential
reloads), and reverses the sense of the statistic to measure a bad event like
most of the other stats (e.g. we want to minimize all counters).
Signed-off-by: Avi Kivity <avi@qumranet.com>