1) You can't write or run scripts "out of the box", you need to do a security incantation first. A oneoff hurdle, but right at the start of the learning curve.
2) Perhaps because of (1), it's not very dogfooded: on a UNIX system, there are lots of already-existing shell scripts sitting around as parts of the OS. PS is not so rich in examples.
3) If you do actually want to parse data, it's not very good at this.
4) I've not worked out how to save intermediate stages in a pipeline to persistent storage.
5) Hasn't yet developed much of a "user culture", although I think that will come eventually.
Not saying that's a bad thing, just that we (as Unix folks) do things intuitively that we forgot we once not intuitive.
Totally agreed on 2 though 5 though.
You can orchestrate scripts in FP style, similar to the REPLs in Xerox PARC workstations, access any .NET, DLL or COM API as if it was part of Powershell, manipulate data without parsing it all the time, interact with running applications via Windows APIs....