The E value of the dummy element should be equal to the actual material being used. Only the weight density should be 0.
Basically what matters is the stiffness. So, Elasticity as well as area & moment of inertia of the section matters.
The issue is happening purely due to incorrect tendon force transfer. The tendons have been assigned to dummy elements with negligible properties. Now, the relaxation and creep and shrinkage losses would come in accordance to definition for plates.
However, the immediate loss would be calculated in accordance to the section property to which the prestress has been assigned.
Since the section has negligible property, initial loss becomes a very high force. This can be checked very easily in Results> Results Tables> Tendon> Tendon Loss.
In this particular case, I wouldn't recommend going for such modelling. If feasible, kindly create a grillage model rather than using a dummy element approach.
On simply considering actual section property to which the tendons have been assigned, the results are better.