Your workflow will not technically be supported, so yes it is dangerous.
Consider this documentation:
Replication and geodatabase releases (10.2.x)
Geodatabases built using previous versions of ArcGIS do not support
some of the newer functions of ArcGIS. Consequently, if your
organization has geodatabases at different releases, consider the
following when creating replicas:
For all replication types, the child replica geodatabase must be the
same geodatabase release as or a later release than the parent
Caution: If you upgrade the parent replica's geodatabase, you also
must upgrade the child replica's geodatabase before synchronizing.
There are additional scenarios described in that document that are also relevant to your 10.0/10.2 situation.
The risks are pretty high even with simple data, and will almost surely result in errors and potential corruption if you are replicating complex datasets such as topologies, geometric networks, etc. You can potentially corrupt the child geodatabase after only one synch operation.
Depending on your replica type (which is not specified in the question - One-Way versus Two-way, synch direction, full or simple data model, etc.), you may not see any immediate issues. I still do not recommend your workflow.
If you're going to try it anyway, back everything up and run the synch operation manually a few times and trace the results and verify your changes are migrating. Automating an unsupported workflow can break a lot of things very quickly..
The client-release = geodatabase-release relationship that Michael Miles-Stimson refers to is a great recommendation for performing edits and synchronization so long as you stay within the supported replica scenarios outlined in that doc.