Picard (2.1.0): the CollectHsMetrics discrepancy

Hi everyone,
I'm having trouble understanding some results using CollectHSMetrics from
Picard's latest version (2.1.0). It seems like the values i'm getting for
MEAN_BAIT_COVERAGE and MEAN_TARGET_COVERAGE are incoherent,
because from the documentation these values indicate:

• MEAN_BAIT_COVERAGE: The mean coverage of all baits in the experiment.
• MEAN_TARGET_COVERAGE: The mean coverage of targets that received at least coverage depth = 2 at one base.

In my case, when I use CollectHSMetrics, I provide the same interval_list file
for both BAIT and TARGET intervals. So I should always have a MEAN_TARGET_COVERAGE
value equal to or higher than the MEAN_BAIT_COVERAGE value (equal if coverage
depth is >= 2 everywhere, and higher if some bp get a coverage of 0 or 1).
But of all the samples I've tested so far, all of them have a MEAN_TARGET_COVERAGE
value lower than the MEAN_BAIT_COVERAGE value.

Can someone help me figure this out?

BTW, I'm using this command:

$JAVA -jar $PICARD CollectHsMetrics \
    I=$NAME.sorted.bam \
    O=CouvertureHSmetrics.txt \
    BAIT_INTERVALS=$INTERVAL_LIST \
    TARGET_INTERVALS=$INTERVAL_LIST \
    R=$REF_GEN_PATH/GRCh37.fa \
    PER_TARGET_COVERAGE=CouverturePerTargetHSmetrics.txt

(with $INTERVAL_LIST being the same file for BAIT and TARGET)

Thanks