I'm fearing that this would create acoustic resonances inside the arctube, as the ballast operate at HF AC and not at LF square wave AC, causing overheating and lamp explosion.
The lamp is underdriven as it is 35W and the ballast is 20W.

Nice experiment, lights*plus

I don't think it is a HF ballast. Sure, it is an electronic one, but because of the complexity, I would guess for really an LF AC style.

Wouldnt some tiny bipin base CMH lamp fit into the original PAR enclosure ?

I was about to ask. I think that a G8.5 lampholder and lamp would fit in the original enclosure by the looks of it...

There are 2 problems with the par enclosure. The short but well made wires had to be cut from the lamp & socket end. I had to tear (destroy) the metal screw base to get the wires out; and the lamp (which appears to be G8.5 base) is cemented in the center of the reflector. Didn't break that yet.

Get replacement screw base from a CFL. They are normally only crimped in a handfull of small points, quite easy to take off undamaged

So the Philips 25w lamp here, which works for a bit then cycles, is the lamp in dor123's link (my lamp however is made in Mexico not Belgium). It's a HF AC driver, but I believe the spectrum (lower right) appeared a bit more even to me. I'll get the spectrum next time.

Is it HF ? HID balasts are usually 200-ish Hz (they usually choose odd frequencies like 173Hz i think to avoid resonances with mains harmnics)

Well the lamptech.co.uk link above says: Additional advances in the ballast allowed a standard high frequency AC driver to be fitted into the space available. This is succinctly illustrated in the X-Ray shadowgraph.

Beside taking measurements on a fully working lamp, the only thing that may "betray" a HF ballast is only a small capacitor in series with the lamp (in the order of 100nF or so).

Once the current goes back to the electrolytic capacitors, it could be both.

When measuring, you need to have a fully warmed, good lamp connected.
The thing is, many HID ballasts use HF for ignition and initial glow phase (before the electrodes warm up) and only then switch to the LFAC operation. The reason is, the HF allows to generate the HV and sufficient power for the cold electrode phase very easily and the high frequency helps to ignite the lamp even without the use of really a high voltage, so again easing the ballast design.
Doing such switchover is way easier when you may think: The LFAC "asks" to have all control function implemented within a microcontroller or so, so an 8pin IC (SO8 or so). The gate prediver has to be separated IC as well (another SO8), so the complete controller part could well be just those two IC's. The power part is then just the mains rectifier (I would guess for a doubler for the US market, providing the middle point to the lamp return connection) and an LC filter. Operating the halfbridge at the LC resonance may easily ignite the lamp and warm the electrodes, following operation in PWM may then use that filter to just reduce the current ripple.
So in total only very few components (that counts here) even for such seemingly complex functionality.

For the eventual HF operation: Even so the heneral rule is to not operate the HID's on HF (because the resonance frequencies may meet the ballast frequency), don't forget here the ballast and lamp combination is fixed fro the complete life. So the ballast frequency could be tuned so, it fits into some gap between two resonance peaks of the lamp. Of course, the lamp maker has to make sure, they do not shift in any significant manner.
And the last is is, what I see as more difficult than making the simple HW circuit operate in more complex manner, so why the LFAC I see as more plausible.

And other note: The circuit here (the GE) is clearly different and more complex (although that may be just my wrong judgement) than the Philips one on the Xray photo...