Let us know how this goes in practice.
“Elephant’s foot” is a delightful name for the effect you’re describing. I call it “compression” and “bleed”, and documented it with a rough diagram here.
There are a few challenges to reducing the amount of bleed (or elephant’s foot) that “oozes” out onto the build platform around the footprint of your part. First off, keep in mind the first few layers are overcured with multiple laser passes in order to gain sufficient adhesion to the build platform. You’re correct this (among other things) multiplies the amount of “stray laser light” that accumulates around the edges.
The build platform is also “pushed” harder against the PDMS for the first few layers, at least for the Form 2, for the same reason. I’m not sure if the Form 3 has a similar mechanism (e.g. I hypothesize it might overtension the flexible tank membrane at the start of the print or cause the tank to be pushed down harder against the rollers of the LPU, but haven’t seen any official remarks about this from Formlabs).
On the Form 2, I’ve found the best way to reduce bleed is to adjust the Z Fine Tuning of the printer until you get a) results that are the correct dimensions in the Z axis, and b) prints that adhere “just enough” to the build platform (strong enough to stay attached, but loose enough that you can pop them off without much resistance). I’ve found (a) and (b) tend to intersect nicely, and I’ve managed to all but eliminate bleed for many kinds of direct-on-base prints.
Unfortunately the Form 3 has no such adjustment yet. 
I gather you want to counteract the effect through proactive design tweaks by chamfering the edges of your model where it contacts the build platform. E.g:
| |
| model |
\ /
\ /
________\___/________
| build |
| platform |
|_____________________|
The problem with a model like one above, is that the first few layers do not share the same footprint. The area under the point of the model will be affixed solidly to the build platform. But as you move up through the next few layers and the footprint expands, the edges won’t have a good base below them to “grab” onto. When I’ve tried this, I get crumbling around the edges and a worse result than a bit of bleed.
That’s why Preform has an “Early Layer Merge” setting under Advanced Settings when you generate supports (or in your case, when you generate the extra base thickness). It merges the first few layers of your model to create an aggregate silhouette, then extrudes that silhouette toward the base. You’ll need to play with that parameter (along with Z-Compression Correction) to keep Preform from undoing what you’re trying to accomplish.
The result might look something like this:
| |
| model |
\ /
| |
_______|_____|________
| build |
| platform |
|_____________________|
That gives cleaner edges, and cooperates with the printer’s behavior of overcuring the first few layers.
It’ll take some experimentation to determine the ideal Z-Compression Correction and Early Layer Merge for your part. When doing these sort of experiments, I often model it all outside Preform for fine-grained control, then plop my STL onto the build platform and skip supports generation altogether. But that might just be out of old habit (since I began doing this before Preform exposed those options).
Hope this info helps!