sample times reported by JMeter seem to scale incorrectly with the number of threads

Hello everyone

tried searching this mailing list for a bit, couldn't find a relevant answer
Using JMeter 2.11 (20140227) to load test our Mirth Connect installation
(HL7 frontend)

Got a simple test plan with a thread group that sends a single request to
Mirth via a TCP sampler, 10 times in a loop
Using both 'view results in table' standard listener and 'response times
over time' plugin from jmeter-plugins to check the results
Max connections on Mirth side is set to way more than 1 or 3, HL7 message
ids are generated randomly for each sample, so there're no collisions there
(and they're sent using different sockets anyway)

1 thread:
JMeter side: table <http://imgbox.com/iFc089tn> and chart
<http://imgbox.com/CbD2GoPH>
Mirth side: http://imgbox.com/1cgAgi0g <http://imgbox.com/1cgAgi0g>

response times line up pretty well so far

now the same test plan, but with 3 threads, ramp-up period 3s:
table <http://imgbox.com/AtsgkBny> chart <http://imgbox.com/i16OCBNf>
mirth <http://imgbox.com/PzD0t4c1>

It's as though JMeter sums up response times from all threads, what's up
with that?

Thanks in advance to whoever might shed light on this