>(b) Mandarin nèi vs. Taiwanese lǎi "inside, home" [...]
>So if your lang historically has such a tendency, or has similar
>phonologies, it might be a start.

I actually have a phoneme /T/ with an unclear historical realization - in
any case velar or uvular, a flap or approximant; or a glottal stop. It
becomes a copy of a following voiceless plosive (like the Japanese moraic
obstruent), causing geminated stops; > /t/ in final position (only dentals
allowed), but a velar nasal before voiced stops (no geminated voiced stops
allowed):
/Tb/ > [Nb] > /mb/, /Tg/ > [Ng]. Exceptions are /Tt/ > [xt], /Td/ > [Gd].
So maybe the change to nasals here counts as a velar equivalent to the
mentioned alternation between /n/ and /l/ in the dental region?

>I've never heard of this type of shift of a glide of a nasal, but Mangbetu
has a 
>series of phonologically labialized labials that are realized as labial trills:
>/pw bw mbw/ -> [B_0 B mB]

Labial trill? Now that's a funny consonant... :-)
But maybe I could have labial + /w/ changing into labial + bilabial flap.
That seems pronounceable, unless I'm doing it wrong. Then, similar to the
/T/ > [N] process above, > /pm, bm, fm/ > /pn, bn, fn/.
And I'm sure I'll be able to think of something where [l\] > [n] to complete
the picture.

>I gather what you *want* here is a series of onset clusters of the shapes
>/Tw Tj Tl Tr/, /Kw Kj Kl Kr/, but /Pj Pl Pr Pn/?

Indeed.

>However, there doesn't
>appear to be a reason (or at least, you've told of no such thing) why the
>/Pn/ series would have to continue a former /w/-series

Well, it would be just a cooler way to both simplify the /pw/ series and
produce /pn/ (which I would like to have, reverse-engineered). Putting a
/pn, tn, kn/ series in by hand as one of the original ones and getting rid
of the /pw/ series would work too, of course, but it would't be as elegant,
I think.