Here's a quick one with the 16x08 that got here today.  Orange is the source, below the left channel looped back.  The recorded track is 28 samples early, indicating to me that the reported latency is 28 samples greater than reality.  I'm not gonna futz with this any time soon  :)
 

 
This shows the defaults when I set it to 48K, 64 sample buffer

 

Ya 28 samples is 1/2 ms @ 48 
Good to know the new Tascam drivers are improving. 
 

Johnny, I finally got some time to try this.
For the test, I used Sonar Professional 32 bit, AKAI EIE Pro (ASIO) set to 128 sample buffer size.
It looks like the AKAI is coming in 10 samples early. I can live with that.



I wasn't sure what to expect, because this is a relatively inexpensive unit, but I like this thing more every day.

While I had the loopback hooked up, I thought I would check for coloration from the circuits.

 
I imported a pink noise wav file to track 1.
Track 2 is recorded from the loopback into input 3. 
Voxengo SPAN is inserted into the FX bin of each track(yes, it's the old SPAN. I like it better).
The source file wiggles a bit, especially in the low end. But from 50Hz up. the recording is a near-perfect duplicate.
This is a closed-circuit test, so monitors and room acoustics do not come into play.
Maybe more nerd stuff, but in the spirit of the thread, I gave it a whirl.

Good stuff Bob. You just made me think of another Nerd test , to record the loop back and flip the phase to see if it nulls? Wouldn't that tell us if the A/D D/A etc were adding colour to the signal? 
 
Are you sure that's only 10ms? It looks like more to me but I guess It depends on how far in you zoom. With the Scarllet no matter how far in I zoomed they still lined up as example. 

I'm zoomed in pretty far, Johnny. The labels on the timeline are in 100 sample increments. Open the image in another tab( I've done that with yours as well). They are much easier to see, especially with my old eyes.
We are looking at the standard kick drum from Session Drummer3.
 
That's not 10ms, it's 10 samples. 10ms would be 441 samples, if my math is correct.
This tells me that the EIE Pro drivers are feeding Sonar accurate numbers and Sonar is making the appropriate adjustments.
 
Also, if you wouldn't mind, post your Driver Settings screen(s).
Looking at BobF's screen and mine, I'm seeing roundtrip latency that differs drastically from the Effective Latency.
The differences in Bob's numbers and mine are fairly linear based on the buffer size.
This makes me wonder, when people claim they can hear 3 or 4ms, what are they really hearing?
My own experience has been that the perception of latency is around 19ms.
 
I'll try the null test when I get a chance. Good idea.
I wish more people would get involved here. This is fun, and I think we might be looking at something subtle, but significant.

My findings where that those setting had no bearing on the test results. I with each device I tested at the lowest buffer and the highest buffer setting and both were the same results. 
I posted a screen shot of my Scarlett settings in # 4 
 
Here's the Tascam

This thread is interesting. I wanted to post quick to it so I remember to try it later. I have not noticed issues with my Saffire but am now rather interested to see a number. This would actually be an interesting statistic to include with an interface as it would reflect maturity of the drivers IMO.

Cactus Music
My findings where that those setting had no bearing on the test results. I with each device I tested at the lowest buffer and the highest buffer setting and both were the same results. 

Copy that. And my last post may be straying from the original premise.
What I was looking for is the Driver Settings screen. This show how the reported latency shown in the Sync and Caching screen is derived.
The balance between Input and Output latency is what has me curious, along with your original question.
Both BobF and myself are showing drivers heavily weighted on output latency.
It makes sense to me, and I wonder how much this balance affects the stability of a certain interface with Sonar.
 
mettelus, cool. Looking forward to seeing the Saffire in action.
 
Kids, you CAN do this at home. Let's see what we've got.
 

Okay, I ran this a few ways since the Saffire has an internal loopback. Session Drummer 3 track, 6 bars of stock midi (3 loops) then frozen. X3e, Saffire PRO 24 DSP, 44.1/24, 128 buffer, 319 sample latency (reported).
 
I had to run a second test because initially my preferences had "Use ASIO reported latency" unchecked (manual offset was 0)... when I checked this, I got a pop-up warning that my output was set to none (rather odd).
 
There was a very nice discreet sample in this loop that I used for reference. Here is its placement for each test:
Original - 69467
Internal Loopback (use reported latency unchecked) - 69664 - late by 197 samples
External Loopback (use reported latency unchecked) - 69734 - late by 267 samples
Internal Loopback (use reported latency) - 69473 - late by 6 samples
External Loopback (use reported latency) - 69543 - late by 76 samples
Reported latency - 319 samples
 
Because of the first two tests being so late I checked for possible manual offset when I noticed that the "Use ASIO reported latency" was unchecked. Not sure why/how but it added another data point to the test.
 
I want to look at this in more detail later, but wanted to get the data recorded before bed at least.