Frank,
Well, if a 3458A and 5700, artifact-adjusted to the same traceable standards do agree in all their compared ranges and measurements within the specified limits, there may be a good point in arguing this is good enough, I cannot judge however. I could not explain why this is ok (or not). I read a doc of the US DOD (or a subco) who did an analysis some time ago and it seems it is ok for them, in a nutshell, that a self-adjusted 3458A is verified only every second time. But not always. Same may (or may not) apply for the 5700. Now based on that, where is the limit of what is acceptable in such a 'bootsrap' approach? Always? Every second time? Never? I guess from an ISO9000 perspective it would be hard to explain to the certifying body that two self-adjusting machines agreeing within to be calculated limits (and never verified by a an independent method) are good enough (unless they accept the Fluke paper as-is, but which also claims a 2 year external cal cycle, mostly for AC though) that each of them meets its spec; (another story is the specific 5720 history, built from parts. We should leave this aside here, but people servicing these units have warned me to buy mixed units combined of boards of unknown version configurations, which sometimes are arround. I can not judge this statement).
On the other hand, since the 752 and SR1010 are so simple and follow simple math in their self-cal / math models, this is easily traceable and accepted practice therefore.