It would increase the back and forth required. How about sending mutations like this:
mutation {
set {
<_uid_:0x0a> <pred.rel> <_new_:x> .
<_new_:x> <pred.val> "value" .
<_new_:x> <pred.rel> <_new_:y> .
<_new_:y> <pred.val> "value2" .
}
}
And then we internally automatically generate the UIDs, assign them to the RDFs and store them. We can also return the value of x and y back to the client in the response. This means each call to the server has a good chunk of mutations, avoiding back and forth to generate ids.