Finally there are some data to be published:
Google Docs - Formidon Young Loot Test
This spreadsheet records the data of 4 runs at Formidon valley, with total of 280 mobs killed. Each individual loot is recorded. Some statistics is provided with the first datasheet, along with empirical costs to kill. While I leave the detailed statistical analysis to mathematicians, here are some remarks that are relevant to the subject of this thread.
First, skewed distribution of loot is evident. There are distinct loot slots, five of them can be easily identified. That is, no loot, 82-107 pec, 149-181, 209-278, 313-380 etc. each slot has average value and width, for example, for slot 2 it is 165pec +-10%, similar for other slots.
Second, this distribution is mob dependent. Number of no looters is about 30%, while for low HP mobs it is about 50% usually. The results for ambu show that MA implements different distributions for different mobs. The spreadsheet has some data for Formidon Mature and Formicacida Weak, with very similar HP - numbers look also very similar, so I think the distribution for low regen mobs is very likely closely related to HP.
Third, the average return without globals and good minics is a little bit less than 50% of the avg. cost to kill a mob. That is very close to what I get on bad hunting and mining runs.
Why this distribution exists at all? Because to have fun of globals and hofs MA has to implement a sort of distribution scheme, in which, for example, 50% return is almost guaranteed, 85% is quite likely with some minics and globals, and the rest is payed as hofs to fortunate ones. It is possible (MA side) to predefine the probabilities of individual loots in such a way that the total payout never exceeds the theoretical minimum of expenses and MA never has to worry that it pays more than it gets.
As a side note - it seems that this scheme is implemented not based on static probabilities, but as a time-based scheme, as it is a common feeling that globals and minics come in batches. Some people interpret it as a "reimbursement" for their expenses, but personally I consider this idea very misleading (and encouraging for more losses) and don't see any grounds for it. Expenses can be raised very easily - shot 10k ammo in the open air or drop 100 bombs at PA - what next?
Now some guesswork. My feeling is that the variance of +-10% of loot slots is not entirely casual. It accounts for three effects: first is pre-defined statistical variance, second is the difference in the damage done by the hunter to the mob. Jimmy's tests will prove it as usual beyond any doubt, I believe.
The third effect is the damage that the mob inflicts to the hunter. It is also likely that MA tracks this and pays for this, too. Difficult to prove, but there are some indications for this. First, there are mobs that have same HP but different damage. Second, some observation: during the fourth run I got a lagged mob that ran forth and back unable to hit me (and unable to regenerate, I think). I killed it finally and got extremely unusual low loot - 82 pec. Never before I had this low loot. Then I paid attention to the damage that a mob inflicts to the hunter and there came the next mob that hit me only 2 times, for 9 dmg amd 23 dmg (rest absorbed by the armor) and gave loot of 84 pecs, the second minimum. Usually formidons hit me 4-6 times for 1-40 dmg average and at the end I have little HP left if I don't fap in the middle.
To summarize: MA know the theoretical cost to kill a mob of a certain HP and pays for this. It is based on HP taken or damage inflicted (the difference is that the last option makes the concept of overkill void). It is possible that damage to the hunter is compensated, too. It is up to hunter to be as efficient as possible to minimize costs and get closer to the theoretical cost (misses, weapon markup, armor and fap decay are his enemies). The payout is based on some dynamic distribution schemes, in which half of the return is allocated to low loots and another half to high loots.
Hopefully further tests will show that my guesswork is total bs and present some "ultimate" evidence how it works lol. Loot discussion is everlasting, have fun