I've had thoughts along these lines as well. My basic reasoning is to keep Evo clean, small and quick.
Actually, when a DBAPI was first being considered, before everything went Revo, I was pushing for this library http://justinvincent.com/ezsql
I'm also considering using PDO for all database activity, but on the other hand I'm trying to get away from resource-hungry code. It may be all well and good for the developers of applications that will be compiled once to native code, but in the case of PHP web apps it's mostly interpreted at run-time. There's a trade-off between ease of development (at any rate as long as what you want to do is what the original developers had in mind) and performance. Besides speed questions, memory usage is enormously inflated with heavy OOP.
Template variables need to be completely re-worked as well, so they are more integrated with the core than bolted on as an afterthought as they are now. Excellent afterthought, but still they can work better. I see them as having one default type, a simple input field, with a much more flexible custom type mechanism. Of course, an installation would come with the basic "custom" types by default.
I'm also considering modifications to the user management system. Nothing like that of Revo, of course (as if I could begin to understand it well enough to even know where to begin!). Actually I'm not so sure that having two distinct types of users is such a bad thing, but I think the user type could better be a field in the main "users" table or perhaps a secondary index table, instead of requiring two complete sets of tables.
Anyway, I promised myself six month with Revo to get to understand how it works. Three more to go. Then I'll have a better idea of what neat features I'd like to backport into Evo. Like property sets. Those are nice, and would be easy enough to implement. Also static elements; again easy to implement and better than having to have "include" snippets.
In any case, something has to be done with Evo; it's been more than a week since the security vulnerability was announced, along with a patch, but no sign of a patched update version release. That's embarrassing.