Admittedly, I don't have much experience with this, but I believe this is how it works, and maybe someone with more familiarity with it will comment:
You check out from a versioned SDE database, and add a feature. The feature has (as well as all checked out editable features) a new UID called AXF_ObjectID (you can see these in the Feature Tables in ArcPad Studio) which performs the function of a UID while the data is checked out. The existing ObjectID will either remain the same, or be blank in the case of a new feature. When the edits are checked back into your version, the features are given new ObjectIDs that mesh with the ObjectIDs in that version. Subsequently checking in that version into the base SDE version will of course compare those ObjectIDs to the existing ObjectIDs where you have to edit them via the reconciliation process as usual if you have duplicates. (I don't know that you have to edit them, the reconciliation process might be aware enough to figure it out on its own, I don't work with SDEs much.)
I believe ESRI recommends a seperate version for each ArcPad checkout you want to have simultaneously so that the ObjectID issue is handled through the conflict detection process.
Hope this helps,
Eric
