I found it very easy to reproduce Howardlong's video.
The phenomenon is there, at all signal levels and frequency / span / RBW / variations.
I used a Rigol DG1032z as a source. Gen and DSA815 were both locked to the same 10MHz reference.
I must say, in practice I don't have much problems with this phenomenon.
Recently I did a lot of measuring frequency curves of 10MHz transformers.
I noticed the amplitude discontinuity at 10 MHz, but with a new normalise it would disappear.
The problem was worst at a TG level of 0dBm. Taking the TG level a few dBm down made the problem much less bothersome.
Here is an example, comparing frequency curves of three transformers:
Regards, Gertjan.