What does it mean to have a "clean" 10 MHz reference in the first place? Low phase noise/jitter or super stable 10 Mhz wrt to temperature/aging etc?
Phase noise is usually not the primary concern, because the usual use of the 10 MHz reference input of an instrument is as reference to a PLL generating some other frequency. And this PLL will tend to dramatically attenuate any jitter at jitter frequencies above a few kHz.
Stability against aging and temperature thus does tend to be the critical parameter.
If it is the latter case: Why would it matter? Say aging: If my setup is stable in short timespans of measurements (say, hours), why would I care if the reference is 10.00000 Mhz or 10.00001 Mhz?
In your measurement it might not matter.
If you want to reproduce your result a year later, it might matter.
If you have a requirement for a particular frequency accuracy, and it's been more than a few hours or days since your instrument was calibrated, it might matter.
I assume aging is not an issue since and I do not care the exact value of the 10 MHz.
If your measurement is not sensitive to small errors in the reference frequency, then the accuracy of the frequency reference might not be critical to you.
Looking at the "Achievable initial calibration accuracy" of the FSW versus frequency error of the SMW and Aging/temperature for SMF (the only given spec) I would go for the FSW (5e-9)
If you haven't recently sent your instrument for calibration according to the manufacturer's recommendations for achieving this spec, this spec is irrelevant.
All else being equal, I might pic the instrument that was calibrated most recently to use as the reference.
But since you implied that frequency accuracy is not critical to your measurement, the whole question is probably moot.
Best Answer
The 'Best Way (TM)' can vary with what you are trying to achieve. Here are some considerations. I'm not going to use the term 'daisychain', as it can mean different things to different people.
Most 10MHz I/O are designed with roughly 50 ohm output impedance and high input impedance. Most have enough sensitivity to work well with a terminated link driven by 0dBm. Most drive at least 0dBm.
This means point to point links between 2 instruments can be connected with impunity. The far end of the cable will receive clean transitions, whether terminated or not, whether driven with square or sine.
A single Tee'd connection is different however. If terminated, all points on the cable get clean transitions. If left unterminated and driven with squarewave, then only the far end sees nice switching. All other points along the cable see the voltage rise to 50% as the outward wave passes, and it dwells there waiting for the reflection from the high impedance far end to continue up to 100%. This midpoint voltage is the worst possible place to wait, in terms of noise and even possible mis-clocking. A Tee'd connection must be terminated at the far end, with all intermediate nodes high impedance. This is less important if you know you have sinewave drive, and will always use sinewave.
Other considerations. You might want to look at the specifications for the internal standards, and choose the best to be the master. Make sure it's got a separate IN and OUT connection, some devices have a single I/O port. The reason? If you decide later to use a higher quality external reference, then plug it in here, and you don't have to re-cable your system.
Having a Tee on the back of the instruments may make cable identification easier when you are crawling around trying to change connections.
You might find that some inputs are just not sensitive enough for good clocking from some outputs, especially if terminated. If you get problems with one configuration, then try another. Better still, measure the sensitivities and output levels, and actually avoid any dodgy links.
Whatever you do, write down how you've connected them, and note which are high Z and which are terminated inputs. It will save grief when you come to add a new box, or exchange it for one with different ref I/O provisions.
If you run point to point links, then it saves Tees, and it may save thinking about terminations (but see below). If any box is switched off, then everything downstream will (may) not work, which is a good failure! If you run a Tee'd connection, then the system may still work with an intermediate box being off or failed. This box may be degrading the standard without you noticing.
If you run point to point links, each box has the option of passing the input straight to the output, or buffering the reference signal. If it passes it straight on, then electrically it's a Tee, and you will need to terminate the far end for it to work properly. If it buffers the signal, it will add noise, which will usually be at a level irrelevant to your measurements. Should you come to investigate an anomalous system close to carrier noise floor, revisit your reference distribution arrangements to make sure that's not it.