Here's what I'd do, as a dialogue editor at heart myself:
cut the scene across a set of DX A-F tracks or more, if needed, but I shoot for using only A-F
split the 'angles' out according to the noise/hum characteristic this may be something Purcell's book didn't address, and is also why I consider his book to be a great introduction to dialogue editorial, but by no means a complete bible - the meat and potatoes of successful dialogue editorial is that most issues are all about situational context, strategy, and sideways/adaptive/intuitive thinking. In a normal scenario you might split out the OMF onto DX tracks by character or shot angle, but intuitively the first way I would approach the scene you desribe when I roll over it in the OMF is splitting out the angles according to the noise problem.
backfill accordingly, overlapping noise is OK the idea here is to smooth everything over gently, especially if the noise bumps in timbre between angles (aside from the harmonic issue). If this means that sometimes the backfill on two different tracks is running simultaneously at times, then there's nothing wrong with that - it's all about making a smooth-playing dialogue track. Editing like this may take a little more time, but it pays off at the stage because when the reel is tossed up, even with the noise issues, it 'plays' and allows the mixer to focus on specific noise issues like the hum without running the risk of disrupting the flow (or having the find/create the flow!)
Notch EQ on the DX track-level << this is particularly if you're doing your own predub, they way I like to predub my dialogue, otherwise a mixer will likely do this instead or have their own method. This is where "write to all enabled" is a GODSEND. basically, you can select the entire , say, 20 second area of dialogue from fade in to fade out where a certain hum or buzz is, bypass the automation and enable loop playback so you can spend time finding and notching out the noise problem. Then, before enabling automation, hit the "Write to all enabled" command, and you're pasted that EQ notch (even if you did multiple ones inside the plugin) across the ENTIRE selection. This is where having done your split outs according to the noise/hum problem makes this method so easy, because you're allowing a dedicated EQ to each isolated noise problem/track instead of dancing between noise types sharing the same track.
After all of that, I recommend trying some real time broadband noise suppression techniques using multiband processors. Cedar DNS is the king but often prohibitively expensive, but the Waves C4 works exceeding well, and for it's price as a single plugin it's a total steal (I've since been on some major dub stages where I've seen the C4 being used in this way too, so I'm definitely not the only one who is using this method nor the only one having success with it).
This is the rundown on it with a before/after sample:
Check out the screenshot of the session too - that sample had harmonic buzz problems from lights in the room and the RED, but the way those edits are laid out are basically a visual representation of what I said above. SO that should hopefully provide some context for what I said and provide you some tangible results. Good luck!
P.S. I'm not against FFT-based processors like RX or XNoise, they definitely have a place in dialogue cleanup and can be valuable. In my opinion though, I don't prefer them at all as a first line of defense. I prefer to evaluate my dialogue through a proper predub sidechain with something like a C4, and only then determine if I need to apply some FFT noise suppression as a roll over a line of dialogue. Sometimes it's quite magical when gently 'lifting' a teeny bit of noise with something like a C4 can do to your dialogue.