I'm getting ready to replace my Roland Octa-Capture with a new MOTU 896mk3 Hybrid.  After getting a suggestion from Jim Roseberry on setting for the MOTU, I checked what I have now with the Octa-Capture.  The Octa-Capture ASIO buffer is set to   
64 Samples (44100Hz) - what Jim suggested for the MOTU.
 
Sonar Reports:
ASIO Reported Latencies (includes Buffer and Hardware Latencies)
Input 4.8 msec 212 samples
Output 2.8 msec 125 samples
Total round trip 7.6 msec 337 samples.
 
That is less than what I had before.  The only thing that has changed is the new version of Sonar.  The input is what I had before, but the Output was about 6 msec (don't remember exactly but that is close if not the correct amount) as total round trip was about 11 msec. 
 
Has anyone noticed lower latency with the 2016.4 version? 

I'll have to check my Octa-Capture when I get home but I recall having very low latency.
 
May I ask : why the change to MOTU?  My Octa has been amazing... what am I missing out on?
 

My Octa-Capture has not been very reliable, especially the headphone jack.  It went flakey and fortunately it was just a bad solder joint on the circuit board.  Then two weeks ago (Friday) the headphone jack went out on one side during a recording session.  I tried to get a replacement jack but was unsuccessful.  I was able to replace it by "robbing peter to pay paul" from another unit that used a jack that was identical.  I can't afford to have a prime unit that is not reliable.  I got the MOTU 896mk3 Hybrid today and am impressed.  The Octa-Capture will be relegated to backup unit.

Isn't that an excessive number of "extra" samples on top of your buffer size? If I set my card to 64 samples I end up below 200.

That is what Sonar reported.  I haven't changed anything in Sonar.
 
However, I got the MOTU 896mk3 Hybrid today. I connected it to USB as I don't have a T.I. Firewire chipset in my PC.  Its samples are lower.  Here is what Sonar reports for 64 samples/44.1Khz (what Jim Roseberry suggested).
Input 2.7msec 121 Samples
Output 3.3msec 146 Samples
Total Roundtrip 6msec 267 Samples
 
 

Oh must be different for every interface then. Was just surprised the difference was half again as much. But maybe it's not so big for larger buffer sizes. Nice to see your MOTU is better already!