Computer PSU's with PFC do in fact have issues with stepped wave sometimes. The most common symptom is the PSU switching itself off under a bit higher load, even when the load is still within the specification of the PSU and it handles it fine when working on grid power

Some combinations of PSU + UPS are worse than others, i think this have to do with the length of the "zero" parts of the wave, and maybe their length change when the load on the UPS change (which in turn is a function of the load on the PSU)

I can imagine that with a ballast that have active PFC something like this may be happening : When the UPS wave is at "zero", the ballast keeps working off its electrolitic capacitor, but since the ballast is not built to work with wave that is zero for such long time, the capacitor is small and the voltage on it suffice to drop to the point that the ballast is working on the edge, or beyond that then it cannot deliver full power to the lamps (at least through part of the cycle)

For the PSU/UPS combinations, I would rather expect the problem in the UPS not being able to maintain the output voltage at that load (generally it follows the battery voltage, so varies over +/-27% just due to battery voltage range of 8..14V, plus some drops in the transformer due to the current) and the PSU not being able to work at the lowered voltage. The mains is expected to be within +/-5..10%.

For the filter cap and the gap it does not differ that much. The gap should be the same width as the "pulse" to get the same relation of Vrms=Vpeak/sqrt(2) as with the sinewave. That means there is (for 50Hz) 5ms full peak voltage, 5ms gap, 5ms negative full voltage and again 5ms gap. After PFC it means 5ms of double power (so double output current of the PFC stage), 5ms of zero, 5ms of double power and again 5ms of zero, so a 100Hz squre wave between zero and double power. On a sinewave power the PFC output current follows the 100Hz sinewave between zero and double power, so in both cases aboutthe same requirements for the filter capacitor.
With NPF inputrectifier the conduction angle is usually about 20..30% on sinewave, while the modified sine gives 50%, so in fact less demand for the filtering capacitor.
So I don't thing that would be the problem.

Many cheap UPS may try to widen the pulse width as the battery voltage goes down in an attempt to maintain the rms voltage, but that does not changes the top voltage, so the level charging the filter capacitor. What it may do is to hide the dropping voltage from your meter when you are attempting to check it, but the electronic really see it dropping...

PSU's with PFC can work on 100..240V and a few still go as low as 60..80V. So the PSU can keep working even when the UPS voltage becomes really low, i dont think this is what cause the PSU to shut down

With the PFC i can see how it makes the coduction angle wider, so the capacitor does not have to bridge so big gap (not 80% of the time as in NPF). so they can put in smaller capacitor. If now the capacitor isnt enough even for 50%, i can see the problem....

This topic relates closely to one project of mine. Ive been thinknig somekind of emergency light system for my "server room" and home-office. Both locations have a separate UPS and simpliest way would be just connect relay to lamp which disconnect it from UPS when mains are presence. I have some unused 18W fluorescent strip lights but all of those are equipped with magnetic ballast. Would it be better to get one with electronic ballast?

Ive been also dreaming that one day I would be able to build -48V DC UPS system. Some of my IT equipments and ventilation system in "server-room" is operating with 48V dc but currently I have to use smaller AC/DC converter without any "direct battery systen". I allready have Delta Energy systems DPS 600B-48-4 with PSC1 controller but its missing rectifier modules and I have not been able to find suitable modules. And if one day I would have 48V DC system it would be cool to integrate backup lighting to that system. Im not sure if there are elctronic fluorescent lamp ballast to 48V dc (46~56v) but some LED exit lights have very wide range in operating voltage.

Interesting. In my case I'm running all mine off a 2000w inverter though so size may make a difference.  All I know is that I've had pretty good success with many different types of fluorescents, from spiral CFLs all the way up to an 8ft F96T12 slimline.  Same goes for running lights on generator power.

PSU's with PFC can work on 100..240V and a few still go as low as 60..80V. So the PSU can keep working even when the UPS voltage becomes really low, i dont think this is what cause the PSU to shut down

But that is the same for the electronic ballasts with the PFC on the input, the PFC just does the same thing - boost the voltage to the required level...
I agree, the lighting is not designed for full performance on such low voltage, because such dips are supposed to be just short time and with lights that means just short dip of the brightness.
And the magnetic will completely extinguish on such dip, so way worse than the electronic (new restart, wth HID it means waiting for many minutes)

I have several UPSes of various types (cheap type for home user, high grade interactive and online) and i would try and give you answers, but i only have working battery in one of the cheap type

So far what i tried with it :

 - CFLs (self ballasted electronic) work perfect

 - Magnetic Switch Start 36W work but the starter keeps glowing slightly, over time it may fail or heat up and melt the casing. I didnt test for long enough to see if any of that happens

 - Incandescent works fine (expected....) but also makes some faint buzzing sound from the lamp



If you try the 18W with S10 starter maybe then the starter won't be glowing, then it should be fine. After all, the choke attenuate very well everything thats higher than 50 Hz

PFC capacitor wont like the harmonics - it must be disconnected i think

The choke attenuates a lot, but with higher voltage lamps (F36,...) during the zero voltage sections the current disappears completely without any voltage available for an immediate reignition, it has to wait till the next active pulse (on a sinewave, the current zero crossing happens at the moment, when there is high enough voltage to support the discharge on the opposite polarity, so the lamp may reignite immediately). That means the reignition spike becomes higher, so the glowing and eventually triggering starter.
With low voltage lamps (e.g. the F18T8 or so) the phase shift is so, the current zero cross happens already during the next active pulse, so the reignition happens normally, so that should be almost fine.
The common problem for all inductors is, the area underneath the pulses is lower than with the sinewave (corresponds to about 150..160V instead of 210 on a 230Vrms), so the ballast current becomes way lower than designed. And that could make the lamps more problematic to operate.

And with the power factor correction capacitors or practically any ballast with any capacitive impedance in it: These are really forbidden on any modified sinewave or so. It not only generates high current spikes, but it may excite rather high energy into resonances at higher harmonics and damage practically anything from the involved devices (the UPS, chokes, the capacitor itself,...), so definitely a "no go" for any cap on anything else than a pure sinewave.