Thanks again to all who replied so far. The list of sources contributed by Seishi Karashima is particularly impressive; I have been able to consult some, but not all.

Thus far, unless I am mistaken, it seems that three kinds of reduplicated formations chiefly occur:

1. The ghanāghana type with lengthened ā (or, once, ū) and an intensive meaning. The examples cited by traditional grammarians seem all (?) to be directly derived from a verbal root (even, pace Renou §147, vadāvada?). Renou §87, however, also cites priyāpriya from the Buddhacarita and jihmājihma from Mhv (= Mahāvastu?). Does this distinction of sources suggest the latter subtype to be secondary and modelled on the former?

2. The nava-nava type without lengthened ā, formed from regular adjectives and likewise intensive in meaning.

3. The menāmenam or kacākaci type with lengthened ā, used adverbially and meaning 'from X to X' or 'X against X'.

Some would include repetitions of inflected words, such as dyavi dyavi, as a fourth kind.

In the instances which prompted my query, krūrākrūraiḥ and saumyāsaumyaiḥ (and, a little later in the same text, śubhāśubhaiḥ) are all (substantivized) adjectives. If the same is true of the priyāpriya and jihmājihma cited by Renou, these would seem to be parallel cases. How, in the absence of a living accent system, such intensive formations are to be distinguished from the far more common dvandva compounds of the priya-apriya type is an open question. Only by context, I suppose.

कृतज्ञतया,
Martin