Do you have parameters for the distributed reference design? What's it's performance? It seems difficult to achieve the intended phase shift over the 1:3 frequency range. The specified max. insertion loss of 10 dB seems however huge, assuming a lossless (pure reactive network), 95 percent of the input wave would be reflected.
To replace the distributed load with lumped elements, analyze it's impedance (respectively S11) and try to find an equivalent circuit.