When I run my mercury lamp vertically, the voltage always stops at 137V +-1V. Today I tried to fire it up horizontally, and the voltage stopped at 130V.

Any explanation?

Also the annoying flicker was gone but that makes sense as with vertical arc the electrodes have different temepratures and emmisions...
So I guess it is better to run it horizontally?

When burning horizontally, the arc bow up, so the quartz below get cooler, so the internal pressure, so the voltage is a bit lower.
Generally lamps alone like more vertical burning, as it mean more even temperature distribution, so the coldest spot is a bit hotter, so better efficacy, and the hottest one cooler, so longer life (less premature failures, mostly on MH).
But it is true, vertical burning causes some weak rectification, what is no issue at all on CWA (as the series capacitor block any DC current), on series choke and HX transformer it might cause some temporary flicker, till the emission surfaces do not adopt - the hotter-burning one start's to evaporate the emission layer more quickly, so it get's a bit weaker, so a bit less efficient, compensating the temperature difference.

Thanks for the reply, do you think that adding a large capacitor (1000s of uF) in series with the lamp could couter-effect the rectification problem and get rid of the flickering?

1) I don't think, you will find a 1000uF capacitor rated for current 4A/50Hz. And it would be an aluminium electrolytic, so short life, or it would be rather expensive (if assembled as foil-type capacitor battery). And how to determine the polarity, as it might change due to mechanisms described above.
2) By adding a capacitor in series you decrease the total ballast impedance (it has the opposite phase then an inductor, so it will subtract)

I would wait few hours, till it doesn't stabilize itself.

Other remark: Flicker is sometimes caused not by rectification, but by an arc root "dancing" around the electrode edge - the total arc glow is constant, but the arc position changes.
This phenomenon happen, when on electrode surfaces is no "favorite" spot for the arc root to reside, so it constantly moves - as convection currents tend to be always unstable... Very often it's in the form of 25Hz flicker - when there are two "favorite spots" and each period the arc root is on the another one.
Occurrence of this is linked to one exact lamp position, it might be either vertical or even horizontal.
And how to tackle this, i don't know - except changing the lamp position.

I would wait few hours, till it doesn't stabilize itself.

OK, but doesn't the top electrode get "eaten away" more quickly, losing its emmision material sooner?

The bulb is rated only for horizontal burning? If it's universal, you don't have to care - it is designed for such condition.

Thanks for the reply, do you think that adding a large capacitor (1000s of uF) in series with the lamp could couter-effect the rectification problem and get rid of the flickering?

I'm not sure about choke ballasts, but I have found that 120v auto-transformer NPF (no cap) ballasts will not work properly with any cap size. Lamp seams to fail to warm up properly, or if cap is too big, lamp will warm up and ballasts will overload and fry.

The older lamp catalogs rated mercury lamps with different outputs depending on whether they were burning vertically or horizontally. The lumen output is always higher in the vertical position, though I believe all mercury lamps except the MA glass arc tube lamps are OK to burn horizontally.

I just switched my 125w lamp from horizontal to vertical burning position. It's running off a cheap HX ballast. The flicker is gone and I did notice it is slightly brighter. I would imagine lamp life could be longer for lamps burning vertically as the quartz arc tube won't deteriorate due to the "hot spot" where the bowing arc gets closer to the tube wall.

I'm not sure about choke ballasts, but I have found that 120v auto-transformer NPF (no cap) ballasts will not work properly with any cap size. Lamp seams to fail to warm up properly, or if cap is too big, lamp will warm up and ballasts will overload and fry.

This depend, if the capacitor is on the same tap as the mains input: If the cap is directly parallel to the mains (always with PF corrected choke reactor ballasts), the lamp, nor ballast behavior will in any way depend on the capacitor value. Different story are HX transformer multitap ballasts, where the capacitor is usually on the highest voltage. They are designed so, the inductive part of the primary current (the biggest one) is handled by whole primary, so on highest voltage available, while the input tap is designed to handle the real input power, assuming the PF is near unity. So any capacitor's unbalance would cause lowering the PF on the mains tap, so increase of primary losses. This happen, when the lamp side is open circuity, but in this condition the increased primary dissipation is no issue, because there is no secondary one, so the total dissipation is not higher then during normal lamp burn.

Oh...I was referring to cap in series with lamp on a HX auto-transformer to try to convert to CWA.

The CWA differ way much more, then a cap in series with a lamp on HX.
On HX, the magnetic shunt is made so, it does not saturate with a margin, as it make the main ballasting impedance, so it's saturation would cause uncontrolled current trough lamp.

On CWA the shunt is designed so, it saturates at the required lamp current. It is this saturation, what regulate the current on CWA: The dominant ballasting impedance is the capacitor, while the inductive impedance (has opposite phase then the capacitive) is effectively subtracted from the capacitive one (the absolute value).
It is designed so, the total impedance provide close to the minimum lamp current on minimum mains voltage. When mains voltage rise, the current increase, but the increasing current cause the shunt to saturate, so the reactance of the inductive part lower, so less is subtracted from the dominant capacitive part. And this compensate for the mains voltage rise - that's why CWA changes the lamp current in less percentage, then the mains voltage change, while in reactor and/or HX ballast of typical design (OCV is about twice the lamp working voltage), the lamp current changes in percentage twice as much as the voltage. And this saturating component would be missing in the HX+cap configuration - so it would be again HX ballast...

Cool....thanks for the info