Post #1 is "technical". It gives the results of the tests and most of what I derive from them. People interested in "how it works" might like reading it.

1) Poisoning
200 attacks (4 times 50 = capital city by vet spies, capital by non vet, non capital by vet, non capital by non vet).
All attacks are successful (city size reduced by one) until size 1 (I suppose that people remaining in the city only drink french wine or belgian beer; you can poison the water with 100 more spies if you have them at hand = no result at all, except a number of spies lost = long live french wine and belgian beer).

Losses are as follows:
1-1 (capital-vet): 17/50
1-2 (capital-non vet): 32/50
1-3 (non capital-vet): 17/50
1-4 (non capital-non vet): 24/50

IMO the way it works is:
Probability to be captured =0,33 when attacking with a vet spy (either capital or not);
Probability to be captured =0,67 when attacking a capital with a non vet (reduced to 0,5 if the city is not capital).

2)Sabotaging with a dip (or a spy using method #1 "use her judgment"):
125 attacks (5 times 25 = dips, vet spies-capital, vet spies-non capital, non vet-capital, non vet-non capital).
All attacks are successful = 1 improvement or 1 unit under building is destroyed.
Losses are 100% with dips, 50% with spies and 25% with vet spies (either capital or not).
The method for choosing the improvement to be destroyed is the same for dips and spies: almost in accordance with the list of improvements in the city screen, with a slight random factor.
If there are n improvements inside the city, the probability that City Walls will be destroyed last, by spy #n+1, is 0,67.
This means that it is very wise to attack the city with n+1 dips or spies, if you want the walls to tumble down.

3) Sabotaging with method #2 "give primary target":
800 attacks (8 times 100: vet or non vet spies attacking walls or non walls (other improvements) in capital or non capital city).
Results:
Column 1 is the number of spies captured before sabotaging, column 2 is the number of spies captured after sabotaging, column 3 is the number of spies unharmed.
3-1 (capital, walls, vet): 46, 35, 19
3-2 (capital, walls, non vet): 67, 26, 7
3-3 (capital, non walls, vet): 26, 29, 45
3-4 (capital, non walls, non vet):51, 39, 10
3-5 (non cap., walls, vet): 39, 28, 33
3-6 (non cap., walls, non vet): 45, 44, 11
3-7 (non cap., non walls, vet): 15, 46, 39
3-8 (non cap., non walls,non vet):28, 45, 27

IMO the way it works is as follows (I think ST agrees with most of this):
Spy #1 close to the city = captured or not captured (works like heads and tails, probability p1 that spy gets captured).
If spy #1 is captured, the same happens again and again until one spy manages to destroy the improvement (this is the tricky part of it, since it means that there is a tiny probability that 10 or even 20 spies get caught before the walls tumble down). Then comes box #2 = captured or not captured after successful attempt.

If I name p1=probability to be captured before successful attempt, and p2=probability to be captured after successful attempt, I give the following values
(might differ from the real values, but rather slightly, I hope):

Vet spies
p1=0,5 in case 3-1 (capital, walls), probability halved if non walls, hence p1=0,25 in case 3-3; probability minus 0,1 if non capital, hence p1=0,4 in case 3-5 and
p1=0,15 in case 3-7.
p2=0,5 in all cases except the most risky one: capital, walls (p2=0,7 in case 3-1).

Non vet spies
p1=0,7 in case 3-2(capital, walls, most risky), then minus 0,2 if non capital or non walls, hence p1= 0,5 in cases 3-4 and 3-6, and p1= 0,3 in case 3-8.
p2=0,8 in all cases except the least risky, p2=0,6 in case 3-8.

Total losses (details already given above)
55 non vet spies are unharmed (and promoted to veteran status) after 400 attacks.
136 vet spies are unharmed after the same 400 attacks.

2-1) Poisoning
Only spies can poison the water supply, reducing the city size by 1, until the city reaches size 1. Utter poisoning gives no result (except spies being captured).
There are no unsuccessful attempts, but there are losses: 1/3 when using vet spies (either capital or not), 2/3 with non vet against a capital, 1/2 with non vet against a non capital city.
If your mother has told you never to switch to commie, the cheapest way of turning your awkward greenie spies into clever vet is to poison the water in a foreign non capital city (50% losses, same rate as when sabotaging "use her judgment", see below)

2-2) Sabotaging "use her judgment" (or using a dip)
All attacks are successful (1 improvement or the unit under building is destroyed) but City Walls are down the list (last on the list twice out of 3 attacks).
Hence, it is highly advisable to take a look (and tell the number of improvements inside the city, let us say n) then come back with n+1 spies or dips if you want to make sure that the walls tumble down.
Losses are 100% with dips, 50% with spies and 25% with vet spies.

2-3 Sabotaging "give primary target"
This method cannot be used with dips (they have no choice).
I would not advise to use it with non vet spies either, except when the number (or the value) of the improvements inside the city is high.
For example, attacking a walled capital city with 1 improvement inside will cost you 2 spies on the average if you use method #1 ("use her judgment") and 9 spies if you use method #2 ("...target").
With vet spies, this method #2 is advisable in most cases (good results with much smaller losses).

quote: The method for choosing the improvement to be destroyed is the same for dips and spies: almost in accordance with the list of improvements in the city screen, with a slight random factor.

quote: If there are n improvements inside the city, the probability that City Walls will be destroyed last, by spy #n+1, is 0,67.

I made a table including probabilities for "give primary target".
Blue: probability that spy will be captured before the attempt
gray: probability that spy will be captured after the attempt
 

non-vet		vet
28	63	15	54	no walls	non-capital
45	80	39	46	walls
51	80	26	39	no walls	capital
67	78	46	65	walls

Your table sums up correctly the results of our tests (BTW how do you get such a nice table? Can one build it using Excel and then stick it?).
You have seen what I derive from them. Of course further testing would help us be more secure about the conclusions. Right now I don't care: NO MORE ATTACKS (I shall be a peaceful warlord next wek ).

It's a way I edit the main Post1 in "Info: dip/spies": I use a wysiwyg html editor, then I replace all EOL's (end-of-line-markers) by spaces using Ctrl-H function in Word (standard html ignores EOL's, Apolyton mixes html and EOL's).
Can one build it using Excel and then stick it?).
Generally, there should be some possibility to revert a Excel-table into html-table. I believe such function is built-in into Excel.

Results look strange  through my sabotage-table. Remember how poisoning was clever after a few tests. When I read Grigor's thoughts, I started to think we may do something wrong.  For example, it may depend on the distance from the capital (I tested for six cities at once, then we mixed that with your results) or on the number of improvements remaining in the city...

I think somebody should do following tests one day:

to sabotage in two equal cities positioned	a) very close to the capital b) very far from the capital
to sabotage for two equal cities	a) with 1/3 of move remaining b) with 3 moves remaining I suppose we did all tests with 3 moves remaining: it shoudln't be a reason of our strange results.
to sabotage one improvement at once and then reload immediatelly (to find out if it depends on the number of improvements remaining in the city)
to test "use her judgment" with ALL improvements of civ2 in the city.

quote: Of course further testing would help us be more secure about the conclusions. Right now I don't care: NO MORE ATTACKS (I shall be a peaceful warlord next wek ).

Playing OCC, I recently had cause to sabotage the progress of a rival AI spaceship - something I'd never done previously.

By allowing the spy to use her judgement, I was able to destroy the current production of spaceship parts on a regular basis. Some of the small towns had no other improvements so it was a 100% success rate. The interesting point is that most of the spies escaped and became vets. (A far higher vet rate than the figures I expected)

In your extensive testing did you ever try spaceship sabotage? (I was playing the OCC #23 game on Deity in 2.42.)

------------