This might be a case for analogue as the universal digital audio connection. Just connect the dats analogue outs and record them with 1 multiface. There will be no quality loss since the 16 bit dats are inferior to the multiface.
Really don't be to anal about these things. On top of that then the 32khz files will be no issue either.

Your reply was perfect! It is just that since digital, we became a little obsessed (at least some of us) with perfection...including bit perfect. I have used dat for years and it is 16 bits (44 and 48khz) and 12 bits, non linear with 32 khz. It might be better to do the last one analogue anyway, since I don't expect the src the cover non linear conversion. Or maybe the 32khz outputs 16 bits while it recorded 12 bits. No idea. It sounded fine btw.
The AD converters of dat where noisy enough not to need dither.

Thank you vinark. That's good to know. I think I will do a listening test and compare A/B to see if I can perceive any quality loss worth being fussy about.

Actually I thought that every conversion stage always degrades the quality slightly but i guess for the practical purposes it might not matter, if both DA and AD conversion is done with the good converters. I also thought that the loss might be due to DATs DA conversion rather then due to  RMEs AD conversion.

Lock means the input digital port sees a valid input source, but not synced to the internal clock (it means it could sync if you set it to). Sync means synced lol.

Yes I understand the idea of loss, but loss is not an abstract, but loss of frequency content, distortion or noise. The DATs AD which recorded the music, is that much behind the multiface that there is no more loss, except that what you already lost when recording onto the DAT the first time. And even then the loss was imperceivable IMHO, if recorded at a decent level. Digital gets messy, just as analogue tape if your recording levels are way to low. Tape gets extremely noisy, digital gets quantisation noise, but digital is more forgiving level wise. Quantisation errors are more ugly if you get them, but you will have to record at extreme low levels, extreme, you can try that.

I re-thought over it.

Am not 100% sure whether it's possible to record from DAT in status "lock".
EDIT: because the multiface sees only a valid clock signal, locks to it, so data transfer is possible, but I fear that it comes to a clock drift, jitter and maybe data loss, what you might hear by clicks and pops, as the multiface doesn't synchronize itself as clock slave, but instead of this works with its own clock.

Therefore a new idea, also because of vinarks last remark (time).

Connect each of the DAT's to one multiface.
The multifaces do not need to be clock synchronisation to each other anymore.
Instead of this each multiface is the clock slave of it's connected DAT recorder acting as clock master.
In the driver settings it's not required to choose the same ASIO buffersize as the three tracks do not need to be sample synchron. I would simply for "symmetry/cosmetic" reasons choose the same asio buffersize. You can take a high value of 1024 or more to minimize the cpu/interrupt load during recording.

Now record 1st the tapes in 48 kHz. You can use all 3 dat recorder in parallel as the DAW can access all of them in parallel through this one ASIO driver. And now each of the three DAT/multiface pairs are properly synched to each other.

At the end process all tapes in 32kHz sample rate.

I hope that it works this way and this should honour all requirements,
- digital transfer
- synced connection between each DAT and Multiface connection and
- saving of time by parallelization.

What do you think of that ???

Ok, then what about three Global Record instances (RME DIGICheck)? Would keep processing load as low as possible as it's designed not to consume too many resources (only available for Windows).

It also won't work with separated DAWs - there is still only one ASIO driver with identical buffers that need to be transmitted synchronously.

Analog is the way to go unless you have a TMS use case (seems not). Digitally the ADI-192 DD converter could be used to record up to for asynchronous sources. Knowing how cheap a computer could be that does not need to operate in lowest latency one could also use four computers (one per MF) and get everything grabbed bit-perfect.

Thank you for all your replies. It cleared it a lot for me.
I was suspecting, that I probably will have to set DATs as clock masters for clean transfers but wanted to double check with the people who know more than I do about this.

As MC says it is dirt cheap to build an extra computer which is a few years old but more then sufficient for the job. In fact we have almost all the spare parts already, so the added cost would be close to zero.
I would prefer to do it on a single computer otherwise. Not so much because of the added time (it doesn't really matter for me whether three computers are recording one track each, or one PC three tracks at a time) but merely due to a couple minor inconveniences with filling in the documentation protocols and backing up the data. But really not so big of a deal.

vinark I double checked the difference between two capturing methods (analogue and digital).
I did percept the small difference in terms of clarity, saturation and a slightly different frequency balance when using analogue outputs.
I think it's not really so much because of the AD stage of the RME, but because of the DA converters and phone/line level amplifiers of the DAT players, which are involved  to provide analogue signal to their outputs.
Being  hardware from the 80s-90s I can imagine they probably are not as clean and saturation-free as the ones the modern interfaces like RMEs have.
Strictly speaking we have to archive these recordings without any alteration, as much as possible, even though for most practical purposes the difference would be minimal.

Thank you for all the info and advices.