Note to the OP: I'm a non-programmer who's just trying to share his views and frustrations:
recommendations for solid, flexible architecture that would be 'future-proof' (as far as possible)
Best future-proof version is to have two separate versions. Just see the appeal of many mobile apps with desktop alternatives.
The reality is that even if you create a cross-platform program, you're dealing with two to three audiences. You're never going to satisfy each of these people's requirements.
Linux users are more likely to crave cutting edge designs from your apps and don't mind you sacrificing a bit of stability for a whole lot of well-communicated developer feedback and adaptation of their bug fixes. It also has to be absolutely free, no strings attached.
MS users want stable and having a standard set of features. These people don't care as much for a program having lots of features as long as you focus on the user interface, the synchronization of data from area to area and basically privacy and ease of updates.
Mac users wouldn't mind no features as long as you put all the eye candy on to it and make it look like a native app.
These three userbase are so separated that you won't really get any flexible future proof cross-platform software without sacrificing things like the bulkiness of java. (at least from my un-educated experience)
It's not that you can't satisfy all these three markets with a decent enough app like most cross-platform programs do but forget about it being fool proof or being special if that's the case. You're not just going to be able to optimize each program for each OS unless you focus more on cross-back ups and cross-data and then just use the difference in culture demand to maximize your program's notoriety on each of those OS. Most of your audience won't really mind if one version is less capable than the other. They just want something that's special and only your program can offer.
If anything, a Linux-specific version with power user options can provide excitement for your Windows audience and your Windows audience reaction can make your Mac audience be more patient with their version while also providing the following that would back up your Windows and Linux audience.
Sure it's costlier and more un-realistic but with your goals, it might be much more realistic to do cross-platform but no app has ever really appealed to all crowds that matched with your lofty goals. (Even popular apps like Opera, Firefox, GIMP, etc. etc. are annoying and not seamless in transition.)
features you would like to see
Snappiness, stability and seamless.
Most of the programs you mentioned are bulky and have a confusing user interface compared to what they can do. It's not that they're non-functional but if anything EverNote has shown, good feature + good use of trialware + familiar user interface and you have an audience and a market.
Hell, you have two markets on the same OS. Except Evernote crapped on one of their markets.
You also have to take note that you can't fit everything into your program so audience won't be looking for a suite and won't want a suite. They want your mini-options as beating the competition in looks, design, feedback, plugins, security and simplicity.
The more you think of focusing on fitting all the programs (even if you think light bare bone alternative) instead of providing a simpler user-friendlier alternative with less features the more you're probably heading the way of Chandler.
features you hope that we will AVOID
Confusing format exporting and importing
Sluggishness in favor of features
Traditional look
Lack of ability to hide each section of the toolbar (see Opera shift + F12)
Lack of relationship between the features
Lackluster tutorial (i.e. clicking help)
Design for the sake of looking cool and large like Windows Media Player
Lack of thought in practical usage
Lack of specific goal the program excels at
Poor gap when jumping from beginner features to advanced ones
Poor website design and lack of clear feature list when downloading the app
Lack of basic "necessary" features when released because you were so focused on developing the app to put it through any real life test
practical considerations regarding building a robust application.
Say No to Featuritis
Don't try to fit two things into one (better cross-format than slow format)
Always take into consideration the systray.
Auto-back-ups, auto-back-ups, auto-back-ups
Never put in a feature half-assedly unless it's for the Linux/BSD version.
1 version beats 2 version; 1 1/2 version loses to 2 1 versions