L-C-L vs Pi network

I don't know how you measured your setups, but having a 0.6 dB loss means either having too high input resistance, or too low output resistance, because the maximum you should get (save peaking due to filters) should be 6 dB, unless you're having some active filtering. Also, a quick simulation with proper terminations shows that your phase readings are quite off.

Judging from the drawings, those resistors should be equal, representing the output of the 1st stage and the input to the 2nd stage. With this in mind, your particular filters there are not "phase shifters" in the real sense, in that there is a phase shift somewhere in there, but most probably not what you'd expect. The first one is a 2nd order bandpass (the two L combine into one), the second one is a 3rd order highpass (PI, as you say). But if what you want is a phase shifter then you need an allpass. Here is a quick test in LTspice:

V(a) (black) is for your first filter, it's a low Q bandpass, V(b) (blue) is a highpass with a little bit of passband ripple that's not very visible (~0.16 mdB), and V(c) is an allpass (that covers the passband of the V(b)) which has a 180^o phase shift at ~146 MHz. Or around it, I just rounded up the values for C/2 (C3) and 2C (C4). You can also see that the phases for the first two filter versions are not exactly 90^o.

If you need a 90^o shift, change the frequency of the LC, or use a 1st order active allpass (needs expensive opamps). Or, if you actually need that highpass in addition to the phase shift then this page and other similar ones might have what you need.