> Tangentially, how does your J-less conlang transliterate the "J" sound in
> "Jim" and "John"? I've done so by substituting "Z" [dz] for it in Cl. Ar.

I assume any of /z zj Z d dz dj J\ j\/ should work, by individual preference. If 
there are no voiced coronal obstruents, /j/ might be a closer match than 
voiceless obstruents.

> In the modern tongue a different problem presents itself: diachronically,
> sequences of [diV] or [djV] are supposed to simplify to [dʒV] and [giV],
> [gjV] to [ʒV]. However, what would happen to [di], [di:], [dy] and [dy:]?
> Ought [di] remain [di] or morph to [dʒi]?

Either way is probably fine, they just require different diachronics. By 
palatalization + assibilation [dj] > [J\] > [dZ] you should get [di] > [dZi] too 
(and probably [gi] > [Zi]); but if you rather go for glide fortition [j] > [Z] you'll 
get zhibilants only from the clusters.

> Ought [di:] be analysed as [dii] and pronounced [dʒi]?

Doesn't sound very plausible, but having said that someone will probably smack 
me in the face with ANADEW...

> How about [dy]/[dy:], given that diachronically they derive from sequences
> of [dui(:)/dwi(:)] or [diu(:)/dju(:)]? Has any natlang encountered such a
> situation before (the lengths are phonemically distinguished)?

> Eugene

Old English did not palatalize before rounded front vowels.
[kYn] > [kIn]
[kIn] > [tSIn]
Counterexamples of languages that do, likely exist too. One branch of Slavic 
has done [kwi] > [tsvi] for a first approximation...

John Vertical