First, the disclaimer: I have NOT played with PocketHAL at all yet, so I am completely unqualified to talk about it.
I CAN however make fair comparisons between GD 2.05 and the last non-PocketHAL version of PF.
It's also fair to point out that my PF project was not the type of game that required a great deal of speed, while my current GD project is.
That all being said, here's some honest opinions comparing both...
(1) The overall functionality provided by PF and GD is, in my mind, essentially a wash. It comes down to what your going to do with them which one is superior in terms of functions.
GD offers more in the way of blits, things like effects. There's many more options in that department than PF. But, PF offers pixel shaders, which can be an incredibly powerful tool (I know, you can more or less do the same things with GD because you have direct surface access... it's not the same thing though, if for no other reason than a pixel shader is a more intuitive way of doing certain things).
PF has more in the way of primitive drawing tools (well, the version I put out does anyway

GD has far superior image loaders. This is perhaps my single biggest complaint with PF, and the one thing more than anything else I hope Thierry addresses in the first PF/PH release (I beileve I read that he is, so I am happy!).
Both provide intrinsic clipping and intersect functionality, so there's no need to re-invent the wheel there.
I'd dare say there's very little you couldn't do in one vs. the other, so let's move on...
(2) Ease of use... both are pretty simple to get going with and to work with in general. I don't think anyone will have problems jumping between them, and getting started probably isn't easier in one vs. the other. Having the source for PF is a plus when you are trying to understanding what's going on at first though.
(3) Performance... ok, we could argue all day here, so I will make what I think is a perfectly fair statement... for 90% of what 90% of all develoeprs will do, the performance of each is not the limiting factor, hardware is. And I'm not talking about WindowsXP needing a 1GHz PC to run effectively vs. a 266MHz PC... what I mean is that your performance will be limited by bottlenecks inherit in the current crop of PocketPC's, namely these lousy graphics co-processors that, how shall I say it... BITE MY ASS!... NOT by either library's code.
Might GD get a few extra FPS in some demos? Sure. Might PF win out in a couple? Yes. Might PH make PF even a bit better? Of course.
But, in the end, I think you'll find either library more than sufficient for the vast majority of projects out there. I think performance is the one area you should look at the LEAST when considering these libraries actually. They are both excellent, and 60FPS vs 56FPS isn't the biggest deal in my mind.
In conclusion, from a purely technical standpoint, I don't see either library towering over the other. There are pluses and minuses both ways, but both are completely useable for most tasks equally well, all things considered.
It's actually nice to have a choice like this where you can be assured that either path will lead to success. You can base your decisions on factors other than the technical merits, to a large extent. For instance, I choose GD for my current project for two reasons, and none else: (1) at the time Thierry wasn't going to be supporting PF, and I didn't want to do it AND try and write a game, and (2) GD offered me the ability to release for the desktop as well (I didn't consider PF's GDI-based approach a viable alternative). The image loader thing mattered too, but I've frankly discovered that my complaint in that department had more to do with the way I architected Invasion: Trivia than the image loader itself, so it wasn't a big factor at all.
ADDENDUM: I should point out that I actually relate to the Shadows far more than Kosh, but I couldn't find a good enough picture of a Shadow. Besides, one First One is just about as good as any other I figure...