Are you suggesting that we build only one “fab unit” at each scale, and then move on straight to the next scale? Would that be wise? If not, then the above statement may not hold at least for the first scale. I expect it would be much more expensive and time consuming to build a large number of fab units using macro technology, than to build one or two, and have them self-replicate.

Of course, you could build a single 1st-scale fab unit, have it build a dozen (or as many as you want) 2nd-scale units, and then go from there. But what if it breaks down before you have enough of 2nd-scale units? If it could self-replicate at its own scale, then every time it replicates you get another backup.

I think that a fab that can self-replicate at its own scale is much more useful, and perhaps not that difficult. A robot arm that can move around a (relatively) wide area, in principle, can assemble machinery much larger than itself. A pair of arms would do even better. If it can assemble another arm, and the machinery necessary to build the parts, then you’ve got everything you need. You could even have it extend its own working area similar to how a crane is extended or how a rail track is laid:
http://www.youtube.com/watch?v=blQUZ4DoV_Y
(notice how the machine moves on the same tracks that it is laying down)

On a different note, do you have any thoughts on how we might deliver the feedstock to the SFF machine, considering that the tooltip would probably need to be moving around? Any channels would have to be flexible enough, but also able to carry the feedstock. Or perhaps keep the tip fixed, and move the workbench instead?