We probably want to have a combination of unique ID prefix in wikimedia projects and URLs that can be used by the rest of the (semantic) web.

I know I have discussed this with Markus before, since it was relevant for WDTK. Can't find where this was though. I've poked Markus now, since I want his input.

using different prefixes (Q => Wikidata, C => Commons, Z => Wikiquote)

Nope, not going back to configurable prefixes :) Guess you where not involved yet at the point where they caused to much pain. It has some further drawbacks that you did not list:

• Not globally unique (can have more Wikibase Repo instances than there are letters)
• Item IDs are no longer easily recognizable as such (similar to: if it's 35 characters and starts with a "1", then I can guess it's a BTC address)
• The prefixes will be rather obscure. If I see "Z123", how would I derive "Wikiquote"?

We do of course not need truely global IDs everywhere. And we have plently of locations where changing the current IDs would be a huge pain. As well as needing BC for certain things for forever. Which I hope is enough to establish that changing the existing IDs we have and forcing the change to happen everywhere is a bad idea. Rather than making such a modification, we need to add new support for global IDs, and switch to using that where appropriate. Which still leaves the question of what the global IDs should look like.

using full urls for item ids (www.wikidata.org/entity/Q123)

If you need IDs global to something more than your own Wikibase Repository installation, then I think it makes little sense to just go part of the way and not make it globally unique altogether. So I much preffer this approach over wikidata:Q123. That is also what WDTK already does if I'm not mistaken. Using the half-way approach in public interfaces seems especially bad. I can imagine that it might be slightly better in some specific internal cases, which however does not make it generally suitable.