Dimensionally speaking, the luminosity of a gravitationally radiating binary system, consisting of two objects of mass $M$, separated by $R$, goes as $(M/R)^5$. The timescale of the chirp for such a system goes as $M^{-3} R^4$. (Schutz 1999).
Thus the total energy released goes as $M^2/R$, i.e. it is proportional to the gravitational potential energy of the system.
Because of the $R^{-1}$ dependence, it is basically the mass and radius of the "final" state that determines the energy lost. For the black hole case, the final mass is just less than $2M$ and the final configuration event horizon is $4M$ (with $G=c=1$). So $M^2/R = M$. Thus I would expect the mass-energy released in gravitational waves to be a fixed fraction of the combined mass of the black holes (note that unequal mass black holes will lead to complications).
Looking at the data for the list of mergers this model seems reasonable, with the fixed fraction being about 5%.
Extending this to neutron stars, well the "final" radius is going to depend on the physics of the neutron star material and so will be model dependent. However, that radius will be $>4M$ (i.e. probably several times the Schwarzschild radius). Another way of saying this, is that neutron stars can't get as close together before the merger takes place.
So from that point of view I would expect $<$5% of the combined mass-energy is radiated as gravitational waves.
Observationally, there is no accurate estimate of the final remnant mass for GW170817.
