The 'newer' MF II is a bit quicker than the old MF, therefore - and because the driver can not know which unit is in use - the latency is not perfectly compensated. When using the ADI-8 the latency can not be compensated automatically anyway, because the unit can not know wat you connect to it. And even then it would be forced to use one value, which might be right for analog I/O or digital SPDIF or ADAT - but not for all of them.

Your argument about parallel compression is not correct. You can not use a digital track and a second DA-Compress-AD digital track and mix them together. Simply because something like 0 sample offset doesn't exist. There is always a subsample delay left which can not be compensated or controlled, coloring the overall sound/effect. So if you want to do that you have to also AD-DA the dry track. And then the latency is of no interest anymore.

I wrote that in my reply, don't know how to say it differently.

From what I gather, the driver can't tell which Multiface BOB is connected (MF-I or MF-II).  RME could likely add a setting in the driver control panel to allow the user to select appropriately.

MC's comment about needing to run the dry and compressed signals through the DA/AD loop does make sense.  People generally record at 44/48/88/96K.  The actual AD/DA chips run MUCH faster than that (called oversampling IIRC) and then "down-sample" to the resultant "base clock" - so they will have offsets/delays which end up as sub-sample delays, and cannot be 100% accounted for with the down-sampled 44/48/88/96K rates.  You can get very close (within a sample), but not exact unless you perform the same DA/AD loop to the parallel "dry" signal.

I haven't done a lot of mixed parallel processing from the DAW (I generally do it "up front" during tracking and record the "dry" and "parallel compression" to their own analog inputs - makes it "1st generation analog" - no DA/AD loops, and is completely phase aligned to the dry with complete wet/dry control in the DAW!) - but even if you get a "0.00" timed DA/AD loop, it can still be off when compared to the "dry" by less than a sample and there's nothing you can do to correct it in a DAW (unless there is some sort of delay plug-in that allows sub-sample delays - basically going to the 3rd decimal place!  I don't think that's possible, but I guess a delay that "oversamples" might be able to get closer?).

FWIW - Nuendo does report "0.00" for a Multiface-I DA/AD loop, so the driver seems to be calibrated for the older MF-I box (to the nearest sample).  But then if you add the ADI-8 to the MF via ADAT, you'll be off again as the driver has no idea what external converters you are using (and their associated delays).  This particular issue has come up again and again, and is basically a limitation of ASIO-2 (ASIO only reports one set of I/O offsets to the host IIRC).  As long as the external AD/DA converter is "Slower" than the RME's internal AD/DA chips (as reported by the driver), then Nuendo/Cubase can appropriately align the loop (again, only to the nearest sample).  If the external AD/DA is "quicker", then Nuendo/Cubase can't cope with it as they can't specify negative delays on the "External Inserts"

I think you might want to rethink your expectations here. This "latency" will certainly not cause any audible offset between individual tracks. For it to be measurable, you'd have to make sure every single drum mic's distance from the drum it picks up is precisely the same (and that is not considering crosstalk). 0.34 ms is about 10 cm in mic distance.

Regards
Daniel Fuchs
RME