The short answer is that counting moras is not simply a matter of counting short and long vowels. Understanding moras, actually begins with counting the types of "acceptable" syllables rather than the actual phonetic length of any given syllable. This is apparently what gives speakers of a language its base sense of rhythm, which can then be altered, sometime substantially, by other factors.
There are different levels of language analysis. The bottom level can be thought of as the level of phonetics, which are objective sound values used in words and where sounds are pronounced with an objective length that can be measured by scientific instruments. Above that level are phonemes, which include a "psychological" component agreed on subconsciously by most or all speakers of a language. For instance, the "p" sounds in the English words "pie" and "spy" correspond to two different phonetical realizations of /p/, i.e., [pʰ] and [p], as in [pʰai] and [spai], or in Ancient Greek spelling, φᾷ and σπᾷι; however, they are felt to be the same "sound" or rather the same phoneme (/p/) by English speakers.
The concept of moras is best thought of not as an objective measure of sound length at the phonetic level, but rather as something linguistically based on the idea that each language has a minimum phonological foot that speakers of a language perceive as acceptable for content words and as giving the language its base rhythm. This is more like the phonemic level of language, if I can stretch the normal usage of the word phoneme.
In Ancient Greek, the minimum phonological foot in content words contains two moras. I think this is the same in modern English, where content words must contain at least two moras as well. A one-syllable content word in English or in Ancient Greek cannot consist of only one mora; however, in other languages, such as in Japanese, there is no such requirement.
Even though the syllables in Ancient Greek can vary in actual length at the phonetic level, they were mostly perceived psychologically to contain only two variations in rhythmic length: light or heavy. That is why closed syllables containing long vowels plus a consonant are treated only as heavy and not as a third type of "extra heavy" syllable. In contrast, if you add a closing consonant to a syllable with a short vowel, you change its type from short to heavy. In other words, adding a consonant at the end of a short syllable changes its type and makes it heavy, but does nothing to a syllable that is already heavy because of a long vowel. In the first case, a consonant has no "length" value, but in the second case, the closing consonant seems to equal one mora.
If all this seems strange, note that adding a certain number of consonants at the beginning of a syllable will change its phonetic length, but does not change the type of syllable, no matter how many consonants may appear. I think this is true even in a language like Georgian which has words like gvprtskvni ("you peel us")--beginning with six pronounced consonants--that are still felt to form only one syllable in length, since there is only one vowel.
Also note the the final syllable of a verse of Ancient Greek poetry was always treated as heavy. It could be that verse-final syllables ending in short vowels were actually phonetically lengthened in performance, but it could also be that the actual or even psychological pause at the end of the verse was enough to act like a closing consonant for an otherwise light syllable, again changing its type into a heavy one. Again, the actual phonetic value was not as important as its psychological phonemic value.
If we take the case of the long diphthongs, adding the iota is like adding a consonantal "y" sound, it might change the default phonetic length of the syllable, but does not change its type, just as adding a closing consonant to a syllable with a long syllable does not change its type. Consider how some people in the American south pronounce the word "right." You might write the pronunciation phonetically in Greek as ρᾱιθ. This is phonetically longer than ραιθ (with a short vowel), which is how many other Americans might pronounce it, but is still the same length as how those other Americans would pronounce the word "ride," which ends in a voiced consonant that automatically lengthens the vowel in American English. In all three cases, these are just one syllable words, containing two moras in English, phonetically having perhaps different lengths, but phonemically having the same length.
As you recite poetry in English or in Ancient Greek, you should adjust your phonetic speaking rhythm slightly to bring out the phonemic poetic rhythm. In English, this is governed by which syllables, regardless of type, are stressed. In Ancient Greek, Latin, Sanskrit, and Classical Arabic, the rhythm was governed by the two types of syllables, based on whether they contained one or two moras. In Japanese, the normal length of spoken syllables are ignored, while the underlying moras are exaggerated. In all these languages, the poetic rhythm is based on rhythm that actually underlies or underlay normal speech, but it is more regularized to create its effect.
Why are there just two types of syllables, and why talk about moras as if they always equal the same actual length of time? I think because two light syllables or one "mora" were felt generally to be the equivalent of one heavy syllable with two "moras." This math works in the poetry of Ancient Greek, Latin, Sanskrit, and Classical Arabic and so became a fixed part of linguistic usage. I think there may be some cross-linguistic reality behind this in most languages, but the details of the poetry in other languages and the details of their acceptable phonetic feet are also different.
