[NOTE: I'm probably wrong about this whole argument, luckily pointed out by Lameen Souag in the comments, see the comments for the discussion.]
For a long time, I've been struck by the irregular nature of the reading traditions of the Quran. The Warš tradition for example, is well-known for its loss of the glottal stop, but for some reason, the loss only affects stems with a root-initial ʔ, this leads to a rather mixed picture where yaʔkulu is read as yākulu whereas ar-raʔsu is read as such. This "Awareness of root-shape" that the reading tradition seems to have, must obviously be informed by Arabic linguistic theory, rather than that it reflects a true linguistic use. If a sound law *vʔC > vvC had operated, you would expect the sound law to be blind to the underlying root structure.
Lately, I've been delving into the Classical Arabic Qirāʔāt literature. This is the body of literature that describes the canonical reading traditions of the Quran. And here I went looking for the origin of the artificial-looking loss of the ʔ.
The first person to canonize the seven most prominent reading traditions was ibn Mujāhid in his kitāb al-sabʕah fī qirāʔāt. Reading up on his description of the loss of the glottal stop, I was struck by the fact that his description of the tradition of Warš does not at all line up with the print editions of the Warš Quran.
Ibn Mujāhid says quite explicitly that any pre-consonantal glottal stop, or intervocalic one is dropped, citing the examples: yuʔminūna> yūminūna, wa-yuʔaxxira-kum > wa-yuwaxxirakum,lā yuʔāxiḏu-kum > lā yuwāxiḏukum and yuʔaddi-hī> yuwaddi-hī.
In other words, Ibn Mujāhid's description is formulated and can be understood as a regular sound law, and ar-raʔsu would be expected to simply become ar-rāsu.
Interestingly, the much later author, Ibn al-Jazarī, who canonized three more readings in his al-našr fī al-qirāʔāt al-ʕašr, to make the now accepted ten canonical readings says something quite different. For the pre-consonantal hamzahe says that it only applies to the first root consonant (and the second root consonant of a few selected nouns) citing the following examples: yuʔminūna > yūminūna, wa-yaʔlamūna> wa-yālamūna, wa-yaʔxuḏu > wa-yāxuḏu, wa-muʔminun> wa-mūminun, wa-liqāʔanā ʔti> wa-liqānā ti, wa-l-muʔtafikātu> wa-l-mūtafikātu. And later on, when discussing the intervocalic hamza, he also says there it only affects the firs root consonant, citing the examples yuʔaddi-hī> yuwaddi-hī, wa-yuʔāxiḏu> wa-yuwāxiḏu, wa-yuʔallifu> wa-yuwallifu, wa-muʔajjalan> wa-muwajjalan, wa-muʔaḏḏinun> muwaḏḏinun and wa-l-muʔallifati> wa-l-muwallifati.
The latter interpretation of Ibn al-Jazarī has become the standard interpretation in the printed Qurans that display the Warš reading tradition. So where did this disagreement come from? It seems that it may have come from the fact that the examples that Ibn Mujāhid cites distracted Ibn al-Jazarī (or, more likely, transmitters of the Warš tradition) from the general statement that Ibn Mujāhid made.
Although Ibn Mujāhid says that the loss of the ʔ affects any pre-consonantal or intervocalic ʔ, the only examples he cites are verbs with ʔ as their first root consonant. It seems that by the choice of his examples coupled with the statement wa-mā ʔašbaha ḏālika 'and whatever is like that', caused later readers of his canonization to conclude that 'whatever is like that' only referred to ʔ-initial roots.
The fact that Ibn Mujāhid's description seems to imply a rule vʔC > vvC and vʔv > vw/yv is important, because this seems to be exactly the rule that has operated in most, if not all, of the modern Arabic dialects. Moreover, it seems to be the rule that underlies the dialect on which the Arabic orthography is based. Moreover, the Quranic rhyme seems to also suggest such a rule was operative in the language of the Quran.
Ibn Mujāhid's description of Warš therefore seems to reflect the development, which had likely taken place in the language of the Quran, whereas Ibn al-Jazarī's description is more 'Classicizing', introducing ʔ into the reading tradition where it was not originally intended.