Hooray! Screws-from-bottom is a much more sensible option, it's in my grand plan too... I've never driven my nozzle into the bed. I have an endstop for that. So making the bed all wobbly seems like a terrible compromise for an unused precautionary function! A rigid bed would make life a lot easier I think.
Thus, I have a few ideas you might enjoy! Note they're all just thoughts, don't have any real experience to share yet. Here goes:
Have a look at optical/laser mirror mounts for how you get highly repeatable, vibration-proof angular adjustment: http://www.newport.com/Optical-Mirror-M … ntent.aspx
Aim of the game is:
* Restrain 6 and only 6 degrees of freedom
* Everything is as rigid as is practicable (Solidoodle seemed not to care too much about this)
* Allow for thermal expansion
Given we have a z-stop adjustment, theoretically you only need two screws to level the bed, and the third (probably the back) could be fixed. Even so, I'd keep 3 screws, as it's quite convenient to adjust the last bit of bed level using the bed screws (during a print!) rather than adjusting the endstop and rehoming.
Use metric screws (welcome to the future!). You can then put calibrated (locking?) knobs on the bottom and know exactly how much you've adjusted the height in each point in real units.
Use a hex-head bolt, as you should be able to make a spiffy printed knob that will press-fit in and glue up nicely, and reliably transmit rotation to the bolt (unlike the glue-on threaded jobs I've got on the bottom of my screws at the moment... Sigh!).
If you use M4 screws (0.7mm pitch, unfortunately a bit awkward), there are threaded steel balls on the internet designed for use on Kossel-style delta printer arms for very few $$ - these could be used on the end of the adjustment screws for a nice low-wear, repeatable interface to a countersunk hole, for example...
Support all three screws from rocking sideways to keep the bed laterally stable (unlike the stock one, which resembles seaweed in the surf) - maybe ones with a longer shank would be good, as the plain part of the bolt will be easy to support snugly in a drilled hole.
Best place to put the springs without causing bending of the bed is close to each of the screws. Note in the mirror mounts above, they but a very strong spring right next to the ball. You only need enough pressure to overcome any lifting forces - of which there is not much under normal printing conditions. If you've got an arrangement (say a ball in a shallow countersunk hole) that might pop out with enough lateral force, then a bit more downward pressure is a good thing to hold it in place.
Careful of thermal expansion. Aluminium, with a temp diff of 100C, will expand 0.44mm over 200mm (bed width). So if you're restraining the hot alu plate, make sure you don't over-constrain it (as the above might do) as 0.5mm is pretty significant. Maybe do the C/S hole + ball thing only on the back screw, then just plain balls running on the flat bottom of the bed for the front two, and some kind of tab in the front middle restrained for left/right movement to stop the bed rotating around the rear pivot point (hooray, 6dof!).
Try to attach the glass plate without putting anything under it - try and get it to touch the alu plate directly. Then if you swap plates (assuming they're a consistent thickness) you'll get repeatable height, and the heat transfer will be as even as possible. Maybe for a glass plate that overhangs, jb-weld (higher temp withstanding than regular epoxy) some washers to the bottom and use magnets (magnets on springs??). To locate it laterally (so it doesn't slip around during printing - repeatability of x/y when changing plates is not so much a concern I think) maybe fixed lip (<= glass thickness) to locate along back and one side, and something with a little bit of give you can clamp against the edge on the front and other sides (e.g. lever action from underneath somewhere. Edges of the bed are colder than the middle, so definitely ok for printed parts here. If your plates vary in size, you can then adjust for this. If you re-use the same plate, then it just gets stuffed back in maybe.
And the rigidity of the bed itself is only as good as the rigidity of how it's mounted to the machine... So make sure you implement some kind of bed stabiliser, and get your z-axis nice and backlash-free / repeatable.
tl;dr - great idea, go for it. 
SD3. Mk2b + glass, heated enclosure, GT2 belts, direct drive y shaft, linear bearings, bowden-feed E3D v5 w/ 0.9° stepper
Smoothieboard via Octoprint on RPi