What you wish for is a standalone player (much like the one that exists for Windows and MacOS X, and somewhat for Linux), not a heavily crippled standalone that is Flash Lite. You'd be surprised how big of a change happened between AVM1 (AS 2.0 & earlier, no JIT, almost no natives, prototype approach) and AVM2 (AS 3.0, JIT, native types, heavy low-level optimizations) speed and byte-code wise - in many cases AVM2 runs faster within a browser than browser's own JS interpreter (which is why plenty of AJAX sites use Flash as an off-loading engine for complex calculations like encryption and such, and even more for its outstanding communication abilities). Flash Lite is based on AVM1, and supports roughly 30-40% of it at that (granted it was upgraded mostly on the video part to allow mobile devices to play JT videos for example), which makes it pretty useless for anything else than banners and playing videos - there are old games that will run or somewhat run in it, but unless the game was created specifically for Flash Lite (which is PITA for any Flash Developer even more than being forced to use hacked-OOP, prototype based approach to programming) you are bound to experience lots of difficulties and unneeded slowdowns. Try surfing through MiniClip on your A5 and see what happens when you try to start a game that was not created with Flash Lite in mind.

Even better than having a standalone Flash Player would be to get an official Adobe AIR port for Maemo - while AIR does add some overhead to the player itself, it would bring millions of Flash Developers to the platform with the ability to create a desktop-grade applications within minutes.

In the lack of both, old tricks can be used that were used in pre-AIR times when one needed Flash to access more than simple Web Services - writing a standalone app that is able to utilize netscape plugin architecture and make it serve only one purpose - play .swf files within the Flash Player. I don't know what kind of webkit support is there in QT, but I'm fairly confident that it could be used to access the Flash Plugin (usually, libflashplayer.so, don't have N900 yet to verify) and pass it a target .swf file. As a matter of fact, if the first two solutions don't show up in timely manner, I do plan to write such an application.

GPU support will only be there for video acceleration, at least for now. There is a possibility to be used for the basic 3D transformations that were added to FP 10, but apart from that I wouldn't rise my hopes too much. They've been promising GPU support for the FP 10.0 since they started talking about it, and there is even added wmode `gpu` (which does nothing), but it didn't bring any speed up. Problem with Flash, when it comes to performance, is that it was, and remains to be a vector-based engine, and that's what makes it so slow - there were lots of optimizations over the time, and AVM2 is pretty awesome as far as VM goes, but the graphics engine of it still remains pretty big CPU hog. Now, they've promised that they'll try to speed up vector graphics by tessellating it and sending it to GPU, but if desktop prerelease versions of 10.1 are of any comparison, it didn't speed it up noticeably. The only noticeable benefit of GPU acceleration so far is video playback - while I have no troubles playing 1080p flash videos on my computer, I know people that couldn't play even 720p before 10.1, and they can run even 1080p now - granted 10.1 really is a beta and it breaks lots of other stuff (not as much as the prerelease, but still not for everyday use).

On the other hand, since 10.1 is developed under Open Screen Project, I do hope that performance improvement will come not from the introduction of GPU in the process, but from some low-level optimizations to the AVM2 (specially its memory management) that Adobe will get input from most mobile manufacturers.

Either way, one thing is for sure, Flash Player 10.1 will run better than 9.4, and possibly even the Flash Lite in standalone mode.