Ok, I've been doing some testing with mods to the demo, to see how low level Carthaginian Phalanx compare to the Hastati, Princeps, and Triarii. Method: Set generals to 0 experience, get rid of general units and all others--so the general becomes part of one of the base infantry units, not suicidal cavalry. Use relatively barren winter map from the mods, get rid of scripting. Use three units of phalanx vs. three of Roman (AI). Set each to 1 experience, and 1 armour, 0 weapons. Use 6-7 deep Phalanx formations of 100 men vs. 100 man Roman units. Try all Hastati, Princeps, and Triarii in successive battles. Then switch to Roman side and try Hastati and Princeps as Romans (more on this in a minute.) I ran my phalanxes around a bit to make sure they were near the same level of fatigue of the Roman units, then met in single line battle. All fighting was initially head-to-head. I did not attempt flanking or try to win, I only tried to keep the lines reasonably dressed. All testing performed on "medium.

Results:
1. The Phalanx bested each of the Roman units in every test.
2. The Hastati seemed to hold out a bit longer than the Princeps (I tried a couple of combined H, P, T tests.) But both were soundly beaten, though neither were "crushed." Kill rate was about 1.75 to 1 vs. these units.
3. The Triarii put up a very good fight. Kill rate was about 1.25 to 1 vs. them. Still, they lost as I would expect going head-to-head.
4. The AI failed to use the pila. Even though I held the phalanx still near closure. The AI charged every time (and tried to charge when I was controlling the Roman units and trying to throw pila!) This gets back to the serious defects in the AI's skirmisher/javelin routine. The difference between the charge/attack threshhold and the missile range are insufficient for practical use--certainly for proper AI use. It doesn't make sense for hastati to charge a stationary phalanx until their pila are exhausted.
5. When I used the pila I was able to do considerable damage (10 -20% casualties). Still, my Roman units lost in melee afterwards.
6. For the phalanx I could not tell that "guard" had any effect at all. I saw the same movement and relative kill rates.
7. The Roman units did work some men in the seams and behind. They would wedge between phalanxes. With follow on units this would be devastating to the Carthaginians.
8. When the AI Romans charged the spear wall of the phalanx they rarely suffered more than one or two casualties before slamming into the shields--often the phalanx suffered as many or more casualties taking the charge. It is my strong opinion that attacker casualties should be much higher for the initial push/charge into the front of a spear wall. This should better balance the phalanx strengths with its weaknesses (lack of mobility.)
9. The speed of the engagement was much more reasonable than in the demo. The high valour units (and generals) are probably having a big effect on kill rates--overdone.
10. The AI did a terrible job with the phalanx. It would not form them into a contiguous line, nor did it make them reasonable depth (I had a range of depths set at the start and it would not alter the shallow one.) I had to be careful to engage them individually and head-on for a decent comparison.

Cool stuff:
The graphics are pretty entertaining during this kind of fight where you have plenty of time to spectate. I watched a number of Carthaginians get "healed." They would get knocked to the dirt, then arise a few seconds later. This happened mainly to swordsmen on the flanks.