Do the other harmonic harmonic generator output spigots have excess sideband noise?

Rich, yes. I took measurements of some of the other ports of the HG in the CER a few days ago (measurements attached).

001.TXT: 45 MHz output

002.TXT: 27 MHz output

004.TXT: 135 MHz output

005.TXT: 90 MHz output

007.TXT: 36 MHz output

There's a clue in the data files you attached.  Looking at the plot of the 45 MHz, you see side lobes at +/- 1MHz.  Looking at the plot of 90 MHz, and 135 MHz you see side lobes with the same offset frequency.  If the source of the pollution was coming into the input of the harmonic generator, then the peaks of the side lobes would scale with frequency.  Given that this is not the case, you are looking for something that effects each frequency directly.  In my opinion, that would most likely point to a power supply issue causing phase modulation of an amplifier common to each output spigot (based on the observation that the induced modulation is about the same in each channel and assuming that's the physical topology of the HG).  

There is 50dB of separation between the peak of the carriers and the side lobes, so you can produce a spectrum like that with a pretty small 1 MHz noise hump on a power supply rail.  I would be wanting to look at the HG power supply directly with an RF spectrum analyzer (be sure to AC couple the power supply to the spectrum analyzer or else you will need to buy a new spectrum analyzer, 1000pF would suffice)

I don't know exactly how you are measuring these noise peaks, but I guess it could also be something in the particular spectrum analyzer (if you are using a different analyzer in the shop for example).  Have you looked at a "clean" signal to be sure you don't see it on all observed signals with that particular analyzer?  Sorry if this is simplistic, you may well have already vetted this and wrote about it only to have me forget somewhere.

I now wish I had popped the top on the HG and its power supply.  I have been itching to do that for a while, but missed the opportunity while I was there this weekend.  I'd be super interested to see a photo of the insides if anyone felt like looking...

The power supply is a LIGO low noise unit boxed into a standard chassis. More details here. Possible points of investigations: power cable shielding, power decoupling at the generator side, grounding issues.

I misspoke.  The offset frequency of PM or FM sidebands in a multiplied spectrum remains constant, but the sideband power scales with the multiplication factor.  

9.1MHz carrier (W) with sinusoidal sideband (Wm) before multiplication:
COS[W*t + A*SIN(Wm*t)]

Derivative of argument for angular Frequency:
W + A*Wm*COS(Wm*t)

After frequency multiplication of factor N:
N*W + N*A*Wm*COS(Wm*t)

Sideband frequency unchanged, power scaled by N

That being said, the sideband power is not scaling by N in the data I saw posted.  The carrier could still be amplitude modulated, but I don't see how it could be phase or frequency modulated.  Sorry for my earlier flawed theory on frequency multiplication.

If someone would please figure this out, I could do a much better job of warping theory to fit observation.

Conclusion:  There might be AM being induced on either on the 9.1MHz carrier, or somewhere inside the harmonic generator - for what little that does to help.

The direct RF spectrum of the 9.1MHz shows no side lopes at 1-2MHz offset. This describes the total power including both AM and FM sidebands. Odd harmonics generators are typically squaring up the fundamental and then filter out the desired frequency. So, they should be first order insensitive to AM.

You are right about the limiting Daniel.  Evan saw sidebands on the OCXO input too.  I attach a plot of the 45.5, 91, and 135 MHz spectra frequency shifted for relative comparison (Data taken from earlier post).